Leximancer Manual

Version 4
2011

Manual Version 4

Table of Contents
SECTION 1. INTRODUCTION TO LEXIMANCER ..............................................................4
WHAT IS LEXIMANCER? ............................................................................................4
THEORY: CONTENT ANALYSIS ........................................................................................................8
WHAT IS CONTENT ANALYSIS? ...................................................................................8
Types of Content Analysis .................................................................................9
INTERESTED IN LEARNING MORE ABOUT CONTENT ANALYSIS? ..........................................10
SECTION 2. THE CONCEPT MAP ........................................................................................10
Theory: Concepts and Conceptual Mapping in Leximancer ...............................10
Concept Seed Words ........................................................................................11
Concept Learning ............................................................................................12
The Initial Display ............................................................................................13
Themes ...........................................................................................................14
Concepts .........................................................................................................18
Buttons in the header above the Concept Map .................................................21
The Concept Cloud ..........................................................................................27
REPORT TABS ......................................................................................................29
Themes tab .....................................................................................................29
Concepts tab ...................................................................................................30
Thesaurus .......................................................................................................34
Pathway tab .....................................................................................................36
Query ..............................................................................................................39
Summary tab ...................................................................................................43
SECTION 3: CREATING AN AUTOMATIC/EXPLORATORY MAP .............................44
CREATING AN AUTOMATIC CONCEPT MAP ................................................................................44
Supported File Types .......................................................................................44
Desktop Installations .......................................................................................44
Leximancer Portal Accounts .............................................................................45
CREATING A NEW FOLDER AND PROJECT ...................................................................................47
New Folder ......................................................................................................47
New Project .....................................................................................................48
THE MAIN LEXIMANCER USER INTERFACE: .................................................................................50
Using the Web Crawler .....................................................................................55
SECTION 4: CREATING A MANUALLY ADJUSTED MAP .............................................62
2A. TEXT PROCESSING ..................................................................................................................63
STOPWORD REMOVAL ............................................................................................68
2B. CONCEPT SEEDS SETTINGS ....................................................................................................74
3. THESAURUS GENERATION: .......................................................................................................78
3A. PRACTICAL: CONFIGURING CONCEPT EDITING ..................................................................79
USING TAGS ........................................................................................................86
THE AUTOMATIC SENTIMENT LENS ............................................................................88

Manual Version 4

Configuring Sentiment Lens .............................................................................88

3B. GENERATING THE THESAURUS ...............................................................................................92
CONCEPT PROFILING .............................................................................................96
4. RUN PROJECT: ...........................................................................................................................99
4A. EDITING COMPOUND CONCEPTS ..........................................................................................99
4B. CONCEPT CODING ................................................................................................................109
KILL CONCEPTS AND REQUIRED CONCEPTS .............................................................................111
CLASSIFICATION SETTINGS .........................................................................................................112
4C. PROJECT OUTPUTS ...............................................................................................................116
GENERATING THE CONCEPT MAP .............................................................................................117
TOPICAL VERSUS SOCIAL MAPPING ...........................................................................................118
CONFIGURING THE INSIGHT DASHBOARD REPORT ..................................................................125
DATA EXPORTS ...........................................................................................................................131
SECTION 5: EXAMPLE ADVANCED TECHNIQUES ......................................................138
1. MANUAL CONCEPT SEEDING .................................................................................................139
CONFIGURING MANUAL CONCEPT SEEDING ............................................................................140
2. PROFILING ...............................................................................................................................144
CONFIGURING CONCEPT PROFILING .........................................................................................146
3. PROFILING USING TAG CATEGORIES ....................................................................................148
Discrimination of Categories based on Semantics ..........................................153
5. ANALYSING TRANSCRIPTS .....................................................................................................160
6. ANALYSING SPREADSHEET DATA .........................................................................................167
PRACTICAL: ANALYSING SPREADSHEET DATA .........................................................................167

Manual Version 4

Section 1. Introduction to Leximancer

-

The learning materials in this section are designed to give the new
Leximancer user an introduction to the workings of the program;
This section also identifies some common applications of the
software;
Theoretical background on content analysis is then provided.

What is Leximancer?
Leximancer is a text analytics tool that can be used to analyse the content
of collections of textual documents and to display the extracted
information visually. The information is displayed by means of a
conceptual map that provides a birds eye view of the material,
representing the main concepts contained within the text as well as
information about how they are related:

Manual Version 4

Essentially, this map allows the user to view the conceptual structure of a
body of text, as well as perform a directed search of the documents. The
interactive nature of the map permits the user to explore examples of
concepts, their connections to each other, as well as links to the original
text.

In this way, Leximancer provides a means of quantifying and displaying

the conceptual structure of text, and a means of using this information to
explore interesting conceptual features.

Manual Version 4

Applications of Leximancer
Application

Type of Text

Output Options

Possible
Projects

Basic Text Analysis

Any non-protected
text:

Visual, via the

Concept Map;

Communication
research;

Word Docs, PDF, Report, via the Insight Analysis of speeches

online content
Dashboard;
over time;
(html), .txt, .xml, etc.
Statistical, via
Blog analysis.
Leximancer data
exports.
Coding Open-ended
Surveys

Customer feedback
data;

Statistical, via
Leximancer data
exports;

Call Centre data;

Qualitative research
spreadsheet data.
Site and Archive
Concept Navigation

Electronic content;
Litigation evidence
(electronic form)

Media Analysis

Electronic media
articles

Customer
Communication from
Relationship
customers
Management (CRM)

Academic Research

Any

Link open-ended
questions to
metadata.
Profile concepts
being investigated;
Concept cooccurrence data.

Net Promoter Score

analysis.

Legal e-discovery;
Alternative to
manually maintained
site maps.

Profile of company or Competitor analysis;

issue
Online opinion
analysis.
Current issues and
concerns of
customers

Policy and campaign

development

Concept Map;

History, Literature,
Media Studies,
Sociology, Politics

Statistical output.

Manual Version 4

Employee satisfaction
survey;

If you would like to see Leximancer in action, example maps can be

found under on the Science page of the Leximancer website
https://www.leximancer.com/.

Manual Version 4

Theory: Content Analysis

-

This section of the manual is for those wishing to understand more about the
theoretical underpinnings of Leximancer;
More practical, instructional chapters are to follow.

What is Content Analysis?

Content analysis is a research tool used for determining the presence of
words or concepts in collections of textual documents.

It is used for

breaking down the material into manageable categories and relationships

in order to quantify and analyze text.

Once extracted, these measurements can be used to make valid

inferences about the ideas contained within the text (such as the
presence of propaganda), properties of the writer or speaker (such as his
or her psychological state), the audience to which the material is
presented, or properties of the culture of the time in which the material
was written.

Content analysis is an important research methodology as it can be used

to analyse any form of verbal communication from written to spoken
forms. As text documents tend to exist over long periods of time, the
technique can be used to extract valuable historical and cultural insights.

As content analysis can be performed on numerous forms of data ranging

from political speeches and open-ended interviews to newspaper articles
and historical documents, it is invaluable to many researchers. Such uses
include:

Manual Version 4

historical analysis of political speeches

detecting the existence and level of propaganda
coding surveys that ask open-ended questions
determining the psychological state of the writers
assesses textual content against measures (e.g. censoring)
assess cultural differences in populations.

Types of Content Analysis

In general, approaches to content analysis fall into two major categories:
conceptual analysis and relational analysis.

In conceptual analysis, documents are measured for the presence and

frequency of concepts. Such concepts can be words or phrases, or more
complex definitions, such as collections of words representing each
concept. One of Leximancers main features is that it can automatically
extract its own dictionary of terms for each document set using this
information. That is, it is capable of inferring the concept classes that
are contained within the text, explicitly extracting a thesaurus of terms
for each concept.

This approach also relieves the user of the task of

formulating their own coding scheme.

Relational analysis, by contrast, measures how such identified concepts

are related to each other within the documents. Leximancer measures the
co-occurrence of concepts found within the text, automatically extracts
this information, and represents the information visually for comparison.
By doing so it displays the main relationships between concepts.

One of the strengths of the Leximancer system is that it conducts both

forms of analysis, measuring the presence of defined concepts in the text
as well as how they are interrelated.

Manual Version 4

The following sections describe

9

Leximancers method for extraction of these concepts and their

interrelationships.

Interested in Learning More about Content Analysis?

If you are interested in learning more about the issues relating to content
analysis and the various techniques that are used, we recommend the
following book: Weber, R.P. (1990) Basic Content Analysis. Newbury
Park, Calif.: Sage Publications, 2nd ed.

Section 2. The Concept Map

Theory: Concepts and Conceptual Mapping in Leximancer
Concepts in Leximancer are collections of words that generally travel
together throughout the text.

For example, a concept building may

contain the keywords mill, warrant, tower, collapsed, etc.

These terms are weighted according to how frequently they occur in

sentences containing the concept, compared to how frequently they occur
elsewhere. Sentences are tagged as containing a concept if enough
accumulated evidence is found.

Terms are weighted so the presence of each word in a sentence provides

an appropriate contribution to the accumulated evidence for the presence
of a concept. That is, a sentence (or group of sentences) is only tagged
as containing a concept if the accumulated evidence (the sum of the
weights of the keywords found) is above a set threshold.

Manual Version 4

10

Associated
thesaurus
items

Leximancer
Concept

Aside from detecting the overall presence of a concept in the text, the
concept definitions are also used to determine the frequency of cooccurrence between concepts.

This co-occurrence measure is what is

used to generate the concept map.

Concept Seed Words
In Leximancer, the definition of each concept (i.e. the set of weighted
terms) is automatically learned from the text itself. Concept seed words
represent the starting point for the definition of such concepts, with each
concept definition containing one or more seeds.

They are called seeds as they represent the starting point of the concept,
with more terms being added to the definition through learning.
Occasionally, more appropriate central terms may be discovered, pushing
the seeds away from the centre of the concept definition.

Leximancer automatically identifies concept seeds by looking for words

that most frequently appear in the text. Alternatively, the user can
manually provide seed words.

Manual Version 4

11

Concept Learning

Leximancer begins with a set of seed words, as defined above. During the
learning process, words highly relevant to the seed are continuously
updated, and eventually form a thesaurus of terms for each concept.

Apart from adding highly relevant words to a concept, Leximancer may

also add words that are negatively correlated with the concept (i.e. words
that rarely appear in sentence blocks containing the concept and
frequently appear elsewhere).

The aim of concept learning is to discover clusters of words which, when

taken together as a concept, maximise the relevancy of all the other
words in the document.

The Concept Map

Once Leximancer has run the learning process and developed a list of
concepts contained in the text, and their relationship to each other, the
information is presented via the Concept Map.

Manual Version 4

12

Understanding the Concept Map

The Concept Map is divided into two sections: a visual display of
concepts and their relationships to each other on the left; and report tabs
on the right for interacting with the concept map.

The Initial Display

When the map first opens, the top 50% of concepts are visible on the
map. These are the concepts that appear most frequently in the text, and
those that are most-connected to other concepts on the map.
Use the % Visible Concepts slider (beneath the map) to change the
number of concepts visible on the map. Moving the slider all the way to
the left hides all the concepts, and moving it all the way to the right
reveals all the concepts.

Manual Version 4

13

Themes
The concepts are clustered into higher-level themes when the map is
generated. Concepts that appear together often in the same pieces of
text attract one another strongly, and so tend to settle near one another
in the map space. The themes aid interpretation by grouping the clusters
of concepts, and are shown as coloured circles on the map:

Here, a cluster of
conceptually related
concepts is grouped by the
theme pipework.

The themes are heat-mapped to indicate importance. This means that the
hottest or most important theme appears in red, and the next hottest in
orange, and so on according to the colour wheel.

Manual Version 4

14

When the map first opens, the Theme Size is set to 33%, but you can
move the Theme Size slider beneath the map to adjust the grouping of
concepts on the map. Move the slider to the right to make fewer, broader
themes, and move it to the left to make more, tighter themes:

When the map first opens, the tab on the right presents a Summary of the
Themes. A bar chart ranks the most important themes relative to one
another, and beneath that the concepts visible within each theme are
listed. A list of representative excerpts is included for each theme, so that
you can read some examples quickly to understand how and why the
concepts in that theme appear together in the text. Hover your mouse
over the More button to see the query syntax used to return each excerpt
within a theme, and click the more button to read further examples.
Manual Version 4

15

In the screen shot below, the Pipework theme (shown as a red circle on
the map) contains concepts such as pipework, LPG and gas. Excerpts
linking these concepts are shown in the Theme Summary in the righthand tab:

If you adjust the size of the theme circles using the slider beneath the
map, the Themes Summary updates to represent the new groups you
have created on the map.
You can make all the themes disappear from the map by moving the
Theme Size slider all the way to the left (0%).

If you hover your mouse over a theme circle on the map, the name of that
theme will appear.
Initially, each theme takes its name from the most frequent and
connected concept within that circle.

Manual Version 4

16

You can change the names of the themes if you right click on the map
near the theme name. A list of nearby concepts and themes will appear.
Hover your mouse over the concept or theme of interest to get an option
to Rename it:

You can make all the theme names visible permanently on the map by
clicking the Map Settings (crossed spanner and screwdriver) button in the
header above the map and tick Theme Names Always Visible.

Manual Version 4

17

Concepts
The Concept Map contains the names of the main concepts that occur
within the text. These are shown as grey labels on the map.

Concepts written with an upper case first letter as assumed to represent

name-like (proper noun) concepts. These often include the names of
people or locations, and appear with a capital first letter on the map.
Examples below include LPG and Calor. All other word-like concepts
appear in lower case on the map, and refer to objects, actions and so on.

Concept Display

Regular Word
Concept
Name Concept

The frequencies with which the name- and word-like concepts appear in
the text are also listed separately in the Concepts tab on the right of the
map:

Manual Version 4

18

Name
Conce
pt

Regular
Word
Concept

The brightness of a concepts label reflects its frequency in the text. The
brighter the concept label, the more often the concept is coded in the
text.

Manual Version 4

19

With concept
visibility at 50%,
some concepts do
not appear on the
map. The grey
nodes illustrate
where they would
have appeared.
Map
sliders

You can reveal hidden concepts by moving the % Visible Concepts slider
(underneath the map) to the right. To reveal the most important concepts
in order, move the slider far left then slowly drag the pointer to the right.

Manual Version 4

20

Buttons in the header above the Concept Map

Hover your mouse over a button above the map to see its name.
Starting from the left, they include:

The Concept Cloud button reveals a graphical alternative to the concept

map. The Concept Cloud will be explained in a separate section to follow.

The Switch to Social Network (Gaussian) button causes the concept to be

reclustered using a Gaussian algorithm. Clicking the same button again
returns you to the original view, created using a topical (Linear) clustering
algorithm.

The Center Map button simply centers the map image in the screen
space.

The Rest Map to Original View button returns the map to the way it
looked when it first opened.

The Recluster Map button scatters the concepts randomly in the map
space initially, then uses a clustering algorithm to allow the concepts to
attract one another once more so as to lay the map out on screen.

The Cluster Map button allows the concepts more iterations of attracting
one another to settle in stable locations on the map (without randomising
them first).

Manual Version 4

21

The Map Settings button allows you to change various visual aspects of
the concept map. Clicking the crossed spanner and screwdriver button
above the map opens this interface:

The Map Settings interface allows you to increase the Font Size of labels
on the map, and to change the background colour from white to black.

It also allows you to make the theme names always visible on the map by
ticking the Themes Names Always Visible button.

You can choose whether to show the spanning tree on the map. The
spanning tree appears as a grey network of connections between
concepts (like a spider web) beneath the concept network. It shows the
most-likely connections between concepts (like a road map of highways),
but there are other (less-strong) connections between concepts (like
backstreets).

Manual Version 4

22

Untick the Name, Word or Tag boxes to hide certain types of items on the
map.

If your map contains tags, ticking Tags as themes allows the tags to give
their names to themes if appropriate. Normally tags are not allowed to
give their names to themes.

The map below was created using the map settings shown above:

The Increase and Decrease Map Zoom buttons allow you to zoom your
map view in or out.

The Toggle Pathway Mode button allows you to turn the pathways
facility on and off. Pathway Mode, when enabled, behaves like this: if you

Manual Version 4

23

click on a concept on the map, then select another, Leximancer will

indicate the strongest (lowest-cost) pathway between those two concepts
on the map:

A black line appears on the map to indicate which other concepts might
be bypassed in order to move from the gas to the explosion concepts in
this example. The tab on the right shows the probability of each leg of
the path, and presents an excerpt of text linking the two concepts
involved in each leg.
The pathways are intended to tell stories emerging from the text, and
focus on indirect connections between concepts on the map.

Toggling the pathways mode off allows you to click on more than one
concept on the map without a pathways being drawn between them.

Clicking the same button again returns you to the pathways mode.

The Export Map button allows you to take a picture of the concept map:

Manual Version 4

24

You can choose the file type and resolution of the image.

Note that pop-ups must be enabled in the browser so that the map
image can appear in a new window. The image can then be saved to local
disk.

The Save Map button lets you save the current map configuration (in case
you have changed theme names etc.) so that you can return to it later.
Simply enter a name for the map image, and click OK.

The Load Map button lets you reload a map configuration saved
previously using the Save Map button. Simply select the name of the map
image you require and click Load.

The Concepts Dropdown List on the right above the map is an

alphabetised concept finder. It allows you to locate (and select) a concept
of interest in the map space. This is useful where you know the name of a
concept of interest, but cannot see it on the map easily.

Manual Version 4

25

Manual Version 4

26

The Concept Cloud

The project results can also be presented via the Concept Cloud. This is
an alternative visual display tool, familiar to anyone who has seen a Tag
Cloud on the Internet.
To display the cloud, press the button at the top-left of the Concept Map:

The Concept Cloud, like the concept map, is heat-mapped, in that hot
colours (red, orange) denote the most relevant concepts, and cool colours

Manual Version 4

27

(blue, green), denote the least relevant. The font size of each concepts
label denotes its frequency in the text.

The Concept Cloud is fully interactive. It behaves like the concept map, in
that you can click on a concept (or tag) to select it and see the list of
related concepts in the right-hand tab:

Manual Version 4

28

Report Tabs
The right-hand window contains several report tabs that allow for further
interaction with the Concept Map. Each tab represents a different way to
interact with the map and explore the results.

Themes tab

As mentioned previously, themes are the coloured circles that group

clusters of concepts. The themes are heat-mapped, meaning that hot
colours (red, orange) denote the most important themes, and cool
colours (blue, green), denote those less important.

The Thematic Summary includes a connectivity score to indicate the

relative importance of the themes. Below this the name of the most
important theme, followed by a list of the concepts contained within that
theme. Examples of text containing each of the constituent concepts in
the theme are then listed beneath. Hover your mouse over the more

Manual Version 4

29

button to see the syntax of the query run to return that piece of text.
Also notice the Hits score to the right showing how many text excerpts
matches each query.

Concepts tab

Clicking the Concepts tab displays a list of name-like and word-like

concepts, ranked by their frequency of occurrence in the text. Clicking on
a concept in this list reveals its connections with other concepts:

Manual Version 4

30

This information may also be accessed and displayed visually by clicking

on a concept on the map itself. The brightness of the ray indicates the
strength of relationship (co-occurrence) between the concepts. A ranked
list

of

the

related

name-

and

word-like

concepts

is

displayed

automatically in the Concepts tab on the right:

Manual Version 4

31

From the related concepts lists, you can browse the locations in the
document where concepts co-occur by clicking in the Browse button (the
magnifying glass icon).

Browsing extracts automatically takes you to the Query tab in the righthand panel, and displays instances where the concepts of interest cooccur in the text.

Manual Version 4

32

Click on Full Text to read any extract in context:

From the Query Results, click on Add to Log to add an extract of interest
to the LogBook for export or reporting. When an extract is logged, the
Add to Log button changes to a View Log option. Click View Log to review
the list of extracts in the LogBook, then select Edit to add your own notes
about an excerpt:

Manual Version 4

33

Thesaurus
The Thesaurus tab displays a list of your concepts, the number of
iterations performed by Leximancer on that concept, and a ranked list of
the thesaurus words that define and describe each concept.

Manual Version 4

34

Click on a concept in the alphabetical list on the left to reveal the list of
Leximancer found to be associated with that concept on the right. The
thesaurus list also shows the relevancy weightings associated with each
indicative word. The iterations count (top left) tells you the number of
times the corpus was reread and coded with evolving concept definitions
before a stable classification result was achieved.

You can click on the Evidence button (the magnifying lens icon) to the left
of a thesaurus item to browse text excerpts where that thesaurus term
appears as evidence for a concept of interest:

Manual Version 4

35

Pathway tab
The Pathway tab displays and describes the most likely relationship chain
between two concepts. It allows you navigate the most likely path in
conceptual space from a start concept to an end concept. To create a
pathway you must first enable Pathway Mode using the relevant button
above the map. Then select a concept on the map:

Manual Version 4

36

Now, clicking another concept will illustrate the pathway between them,
along with example text. The relationships between concepts are best
thought of as correlations, though the text segments describing the
relationship may define a direction for cause.

The example below creates a pathway between the tank concept and the
explosion concept:

Manual Version 4

37

When you identify a pathway on the map, a list of text segments

describing the legs of the path appears automatically in the right-hand
panel. This list is like a narrative of text segments that are relevant to the
legs (sections between concepts) of the path.

The start concept appears at the top of the list, and the associated text
segment explains the link between this and the next related concept in
the pathway. This pattern continues until the final text segment linked to
the end concept is listed.
The link between
connections:

Manual Version 4

tank

and

explosion

revealed

the

following

38

Query
We have already discussed one of Leximancers query functions: querying
one concept against another:

But Leximancer allows for increased query specificity. Entering the

following query terms in the query search bar:

Manual Version 4

39

NAME:[concept]
- searches for your specified name concept
WORD:[concept]
- searches for your specified regular word concept
TAG:[file, folder, or tag]
- searches for a pre-defined tag in your data
WTERM:[word]
- searches for a regular keyword (different to a concept)
NTERM:[word]
- searches for a name keyword (different to a concept)
TERM:[word]
- equivalent to (WTERM:[word] OR NTERM:[word])

These terms can be used in conjunction to search for co-occurrences of

any number of specified concepts, tags, and / or keywords. For example,
you could search for excerpts that mention the keyword corroded with
the concepts pipework and explosion by entering this syntax into the
Query box:

Manual Version 4

40

There are also other, more specialized, query searches available:

ITAG:[file, folder, or tag]
- searches for your specified ignored tag. For example, a
spreadsheet column with more than 500 values is classified as an
ignored tag. Although ignored by Leximancer, you are still able to
use it in your query to narrow results.
SECTION:[section number]
- searches for the section number of the text block
There are also some additional new tools to assist with query searches.
These are query parsers that allow for more complex queries:

NAME:[partial concept]*
- adding a star to the end of a concept search will cause the query to
be applied to the letters present and any derivatives. You must

Manual Version 4

41

provide at least one character for this search. NAME:* alone will not
produce results.
- For example, a query of NAME:environment* will search for
environment, environments, environmental, etc.

NAME:[concept] OR NAME:[concept]^2
- searches for either concept specified, but with the latter concept to
be considered more important
- for example, a query of NAME:greenhouse OR NAME:emissions^2
will search for both greenhouse and emissions concepts, but
emissions concepts will be considered more important.

+NAME:[concept] +WORD[concept]
- this is shorthand code for the search NAME:[concept] AND WORD:
[concept]

+NAME[concept] +WTERM:[concept]
- searches for the name concept co-occurring with a keyword

Manual Version 4

42

Summary tab
The Summary tab displays extracts containing the most important
concepts discovered from the text. The list contains characteristic text
segments that illustrate the relationships between key concepts:

When you have finished exploring the Concept Map project tabs, close
the Map Explorer tab using the Close button (X) in the top right-hand
corner.

Manual Version 4

43

Section 3: Creating an Automatic/Exploratory

Map
The aim of this chapter is to give you an overview and introduction to
using Leximancer. In this chapter, you will be performing an exploratory
analysis of a data set comprised of online media about climate change.

Exploratory maps involve minimal input from the user and are the
starting off point for analysis using Leximancer. They are a means of
gaining an overview of your data before adjusting Leximancer settings for
more tailored results. Manually adjusted maps will be explored in the
next chapter.

Creating an Automatic Concept Map

Supported File Types
Currently .doc, .docx, .pdf, .html, .xhtml, .htm, .txt, .xml, rtf, .tsv
and .csv files are supported. If there are other formats of text data you
wish to process, consider pre-converting them into plain text (.txt) first.
Desktop Installations
If you are working with a desktop version of the software, click on the
desktop shortcut, or select Leximancer 4 from your Start menu. The first
time you run Leximancer, you will be prompted to enter your licence key.
If you have downloaded your licence key file, clicking Select Licence Key
allows you to navigate to the location where its saved. Click Open to load
the key information, and then OK to start the program:

Manual Version 4

44

A Leximancer icon will appear in your system tray when the programme is
running. To exit the application, right click on the icon and select Exit
Leximancer.

Leximancer Portal Accounts

If you are using the online Leximancer portal, open your internet browser
(for example Mozilla Firefox or Internet Explorer) and navigate to the
Leximancer 4 url (www.leximancer.com/lexiportal/). Login using your
username and password to start a session.

When the program starts, youre presented with the Manage Projects
interface:

Manual Version 4

45

Two top-level folders are visible: Tutorial Projects (a folder containing

generic example projects that come pre-loaded with the software), and
Leximancer / User Projects (a folder to house your projects). Click on the
plus sign adjacent to a folder to expand the tree and view any folders or
files within. In this case, there is a project called Explosion Enquiry
inside the Leximancer Projects folder:

Double clicking on a projects name opens that project.

Right clicking on a project allows you to Rename or Delete an

existing project.

Right clicking on Leximancer / User Projects, then selecting Create

Project allows you to create your own new project.

Create a hierarchy to organise your projects by housing New

Projects in New Folders.

Manual Version 4

Creating a New Folder and Project

New Folder
Right click on Leximancer Projects (Desktop) folder, or open the User
Projects folder and then the folder by your name (Portal), and select the
New Folder option. You can then name (and optionally describe) the New
Folder.

As a project file or folder is created using this name, try to use typical
conventions for naming files (for instance, avoid including characters
such as ".", "\", "*", or "/").

Manual Version 4

47

After you have named the folder, click OK. This creates an empty project
folder by this name under Leximancer / User Projects.

New Project
Right click on your new folder under Leximancer / User Projects, and
select Create Project to name (and optionally describe) a new project.

It helps to be descriptive when naming your projects, so that later, when

you have multiple projects, you dont need to open them to check their

Manual Version 4

48

content. As we are creating an Automatic Map, I will name the project to

reflect that:

This opens the newly-created project, and displays the main user
interface in the right hand panel:

You can open several Leximancer projects at once. When youve opened
one or more projects, you can also collapse the Project Selection
interface using the arrows:

Manual Version 4

49

Having created a project, you are now ready to use the Main Leximancer User Interface.

The Main Leximancer User Interface:

The user interface is a flow diagram showing Leximancers phases of
processing to generate a conceptual map. The Project Control Panel
consists of a left hand column that gives the name of the processing
stage:

A middle column allowing you to change the settings for each processing
stage:

Manual Version 4

50

And a status column that keeps track of the completion of each stage of
processing:

For an Automatic Map, we only need to complete the Load Data step, and
then click Run Project. Both of these buttons appear in orange in the
main user interface:

Manual Version 4

51

When you click Load Data,

If you are using Leximancer Desktop:

- You will see your local drives listed in the Available Documents panel on
the left.
- Expand the drive directories to see the files and folders within them.
- Drag and drop the desired files and folders into the Document Set area
on the right to link them to the project, then click OK.

If you are using Leximancer Server or Portal:

Manual Version 4

52

- Right click on your user data folder (usually your name or company) in
the Available Documents panel on the left, and select Upload.

- Browse to locate the data on your local machine. If you wish to upload
multiple files at once, place them in a zipped folder before selecting them
for uploading. The files and folders will be extracted from the zipped
archive automatically on upload to Leximancer.

In the File Upload window, select the documents you wish to analyse and
click Add. Once all the documents you want have been selected, click OK
to begin uploading:

Manual Version 4

53

In this case, we have uploaded a single zipped folder containing a

transcript for each day of a hearing after an explosion at a plastics
factory.

There is some feedback in the lower left of screen to let you know that
your files are uploading, and when it is complete.

Once the data is uploaded, you can expand the parent folder to see the
files and folders within:

You could choose to analyse just one of the documents within a folder
(e.g., a single days transcript) by dragging and dropping an individual
file into the Document Set area on the right. Alternatively you can drag
the whole parent folder into the Document Set area on the right to
analyse all its contents.

Selecting the parent folder instructs Leximancer to analyse all the files
and folders within. This allows you to analyse multiple files and folders at

Manual Version 4

54

once, and facilitates comparisons between groups of documents using

the softwares automatic tagging options.

For Sever and Portal Users:

Using the Web Crawler
If you have arranged to use the web crawling facility, after clicking Load
Data, right click on the folder by your name (or your companys name),
and click web crawl:

The following interface will appear:

Manual Version 4

55

Provide a name for the web crawl, then press Add Sources and type in
the Url(s) you are interested in analysing. Blog and review sites tend to
work best.

Tick the box to the right to do a Single Page Fetch if you only want the
text on that page to be retrieved.

In the Advanced Settings you can set the Level of Recursion. This defaults
to 2, allowing the crawling to follow links from the page to 2 levels down.
You can also increase the maximum data size if you want to crawl a large
amount of textual data.

Enter keywords into the Specific or Global Content Pattern boxes to focus
the crawl on pages that mention particular words.

Manual Version 4

56

The web crawl will begin when you click Ok. When it is complete, you can
drag and drop the folder of crawled text into the Document Set area, and
then run the project as usual.

Manual Version 4

57

For All users, once you have moved some files into the Document Set
area:

You have the option at this stage to specify the type of document(s) you
wish to analyse. The current example uses online data, so .html was
selected. Selection of data type helps Leximancer to be more sensitive to
the idiosyncrasies of that data type (for example, html artifacts from
online documents).

You may also specify the language of your documents:

And the type of Character encoding used:

Finally, click OK to link the data to the current project.

Manual Version 4

58

This returns you to the main user interface, where the Status of the Load
Data stage will be green, as this step has now been completed:

To perform an automatic or exploratory analysis, click the Run Project

button in orange at the bottom of the Project Control Panel. This will run
the project using default settings.

While the stages are running, they will flash orange and say In Process.
Once they are completed, the status nodes turn to green in colour and
display the word Ready. Progress information is also visible at the bottom
left of the screen:

Manual Version 4

59

When all phases of processing are complete, either click the Concept Map
button at the bottom on the main Control Panel, or click Open Map in
New Window (in the upper left of screen to open a larger map in a new
window).

Manual Version 4

60

The Main User Interface also includes:

A Close tab button (x) that allows you to close the current project
(current settings are saved) to exit or load other projects;

A Cancel Processing button that stops Leximancer processing,

wherever it is up to at that time;

An option to Import Settings or Export Settings, in which you may save

the settings of your project, or apply settings from an earlier project;

Manual Version 4

61

Section 4: Creating a Manually Adjusted Map

The manual will now discuss the ways in which the user may alter the
settings of their project to create a concept map. This section will follow
the sequence of the stages as represented in the control panel (below).

However, the Load Data stage will not be covered in this chapter. Please
refer to the start of the previous chapter, Creating an Automatic/
Exploratory Map: For Beginner Users (page 38) for information on how to
start Leximancer and load data.

Stages of Processing
1. Load Data:

Manual Version 4

62

Please refer to the previous chapter (page 38) for information on how to
load data in Leximancer.

2. Generate Concept Seeds:

You can run this stage of processing using default settings by

clicking the Generate Concept Seeds button.

Alternatively this stage can be expanded (using the plus sign next to
Show Settings) to reveal two sub-stages within:

There are options to Edit the Text Processing and Concept Seeds Settings.

2a. Text Processing

This is the first phase of processing that is run from Leximancers main
menu. This phase converts the raw documents into a useful format for
processing. Preprocessing involves the following steps:

Splitting the information into sentences, paragraphs and

documents.
mark

These boundaries are important as they generally

transitions

in

meaning.

The

conceptual

map

of

the

documents extracted by Leximancer reflects the co-occurrence of

distinct concepts. To prevent concepts from being perceived to be
related across changes in context (such as across different

Manual Version 4

63

documents), the co-occurrence is only measured within (and NOT

across) blocks typically containing 2 sentences.

Removal of non-lexical and weak semantic information. Within

each sentence, the punctuation is removed along with a collection
of frequently occurring words (called the stop-list) that hold weak
semantic

information

(such

as

the

words

and

and

of).

Furthermore, for documents extracted from Internet email and

news groups, the headers are cleaned up and the non-text
attachments are removed.

Identifying proper names, including multi-word names. Often

in documents the proper names (such as people, places or
company names) depict important entities that should be mapped.
For this reason, proper names are extracted as potential concepts.
In Leximancer, words are classified as proper names if they start
with a capital letter. As every word that starts a sentence falls into
this definition, only start-of-sentence words that are not in the
predefined stop-list are considered as names.

Optional prose test of each sentence.

To remove non-textual

material from the text, such as menus and forms in web pages,
sentences that are unlikely to be part of the specified language are
removed. This is achieved heuristically by removing sentences that
contain less than 1 (or 2) of the stop-list words. If processing
spoken language, this setting should be turned off.

Practical: Configuring Pre-processing in Leximancer

Manual Version 4

64

Clicking on the Edit button for Text Processing reveals the following
interface:

Manual Editing Options:

Sentences per Block (1-100): This option allows you to specify the
number of sentences comprising each context block (or text segment). A
context block is said to contain a concept if the words therein provide
sufficient cumulative evidence of its presence.

Commentary: The best value for this parameter depends on the

nature of the data. It should almost always be two or three, though
in some instances one sentence is sufficient (eg: press release data
or verse)

Prose Test Threshold (0-5): The Prose Test feature examines raw text
sentences to decide whether they are valid prose from the configured
languages. This is achieved by counting the number of stop-words that
appear within each sentence. If this number is high, it is likely to be a

Manual Version 4

65

sentence from a configured language. This option allows you to specify

the number of stop-words that are required for the sentence to be
further processed.

Commentary: This feature is good for reports or other prose

documents where you dont want to process tables of numbers or
lists of words. It is almost essential for web pages or e-mail
messages which often contain menus or signatures. This sort of
repeated data can potentially contaminate your automatic seeds
and machine learning. If the data is not prose from a supported
language, or if it is composed of transcribed speech or other
colloquial matter which does not obey the rules of prose style, then
this feature should be weakened or disabled. You should also
disable this if you need to analyse absolutely every bit of text in the
data

Duplicate Text Sensitivity (Off|Auto|1-8): This setting suppresses the

processing of duplicated text. This option is especially useful when
analysing email data or blogs and reviews, where cross-posting and
quoting is common.

Identify Name-Like Concept (Yes|No): This setting is important if you

would like words that seem to be names (i.e., non-stop words, starting
with a capital letter) to be stored as potential concepts.
This setting requires text data which uses upper and lower case, where
upper case designates proper names. This doesnt work in many
languages, or in some text data where case is missing, but it is very

Manual Version 4

66

useful much of the time for tagging proper names. Note that it binds
compound names into one token.
Break at Paragraph (Yes|No): This setting is to prevent context blocks
from crossing paragraph boundaries. Only if the majority of paragraphs
in the text are shorter than 2 sentences should you consider ignoring
paragraphs.

Auto-Paragraphing (Yes|No):

This setting identifies whitespace,

particularly line-breaks and paragraph-breaks, if the document is

consistent in its spacing, to identify new paragraphs. If there is a
document with no reliable spacing for paragraph boundaries, this setting
should be turned off.

Merge Word Variants (Yes|No):

This option employs a stemming

algorithm to identify the headword for initial thesaurus items. For

instance if stemming is turned, the initial thesaurus terms for the stem
look in the Concept Seed Editor may include looked and looking. If you
don't like the results, you can ungroup the thesaurus items in the seed
editor. Lemmatisation is off by default.
File Preparation (All Files|New Files Only): The New Files Only setting
allows you to optimise processing by preprocessing only the new files in
your data folder (i.e. files that have not already been preprocessed). The
All Files option will delete all previously prepared text, and start the
preprocessing from scratch.
Commentary: The normal situation involves mapping a fixed set of
documents, and in this case you want any previously prepared text
cleared out from the map. In case you want to add some new files
from a data folder to an existing map, then use this setting to stop

Manual Version 4

67

the previously prepared text from being deleted. This will also
prevent the importation and preparation of any files in the data
folder which have been prepared before.

Stopword Removal
During Preprocessing, words with low semantic-content (meaning) are
removed from the text data using a predefined Stopword List. An
example of an English Stopword is 'and'. This word occurs frequent in
English text, but would not constitute a useful or clearly-defined concept.
If you are using an unsupported language, you can update this list (e.g.
by translating the contained words into your language). Note that stop
words are removed from the text to analysed, and cannot be selected as
manual seed words.

Commentary: Stopwords are frequent words in a language that are

rather arbitrarily designated as having little semantic meaning.
Leaving stop words in the data has an obvious effect on the
automatic seed selection. If you leave the stop words in, some will
be chosen as automatic concepts. This can be annoying, depending
on what you are looking for. The presence of stop words also
impacts on the machine learning of thesaurus concepts, since
almost everything can be correlated with words such as is or
and.

Edit Stopword List (Button): You can check the words that are counted
as stop-words by clicking this button:

Manual Version 4

68

The following interface will appear:

You can browse through this alphabetical stoplist and see if you find any
words that you would rather have left in the analysis. If so, click on the
Remove option next to the word, and change the status of the word to
Allow using the dropdown menu. Set the words status to Evidence if you
Manual Version 4

69

want to allow the word to form part of a concept thesaurus, but not to be
considered as a possible concept candidate itself:

You can edit words in the list by clicking on them and typing in the text
box thats revealed. You can also Add words to the stoplist, or Remove
them from the list. The Download button lets your download the
current .xml stoplist file, and the Upload option lets you upload
another .xml stoplist file.

Manual Version 4

70

The Language column lets you know the language from which particular
stopwords come in case you are analysing documents in different
languages. You can change this setting so that stopwords are only
removed

from

text

in

the

appropriate

language.

The

language

abbreviations in the list are ISO 639 country codes.

You may also add the stoplist of another language. This includes highfrequency, non-semantic words from a wide selection of languages. Note
that the relevant stoplist will automatically load after selection of the
language in the earlier document selection window:

Manual Version 4

71

Press OK to return to Pre-process box, then click OK to return to the

Control Panel.

Tagging Options
Tags are important for comparing different documents based on their
conceptual content, for example, different speakers in transcript
documents, or for a comparison between different text sources. At this
stage in processing, you can instruct Leximancer to pay attention to
certain tag categories so that you may analyse them later.

The Apply Folder Tags (Yes|No) and Apply File Tags (Yes|No) options
can cause each part of the folder path to a file, and optionally the
filename itself, to be inserted as a tag on each sentence in the file. In our
example project, we can use the File Tagging facility to compare the
content of the transcripts on different days of the hearing. Source
document tags can then be included as concepts on the map.
Commentary: This is a powerful feature that lets you code all the
sentences in each document with categorical tags just by placing
the files in folders, possibly within other folders etc. The tags can

Manual Version 4

72

then be included on the map, among the topical concepts. This is

useful for performing document clustering in a semantic context,
or for making a discriminant analysis between two categories. For
example, if you had a set of political speeches from a debate on
some issue, you could give each speech file the name of the
politician, and place all the speech files from each political party in
a folder named as the name of the party. Then, if you had several
sets of these speeches from different years, you could place each
years set of folders in its own folder named with the relevant year.
Applying folder and filename tags will then insert the name of each
politician as a tag on each sentence, and the name of the
containing political party as a separate tag on each sentence, and
also the name of each year. When you map this data, you will find a
concept for each politician, each party, and each year in the Tags
collection. You can then choose which of these dimensions to
cluster together on the map, so you could view the issues by year,
by party, by politician, by year and party, by politician and party, by
year and politician, or all of them co-varying at once if you are
adventurous.

The Apply Dialogue Tags (Yes|No) function is designed to utilise

speaker labels or headers in the text data. This setting identifies speaker
labels that start with upper-case, end with a colon and space, and are
located at the start of a line. Each speaker label is appended to the end of
every subsequent sentence until a new label is found. This can be useful
for analysing focus group or interview transcripts.

Manual Version 4

73

2b. Concept Seeds Settings

This the phase of processing in which seed words are identified as the
potential starting points of concepts. Concept seeds represent the
starting point for the definition of concepts. They are single words (such
as violence) that are the potential central keywords of distinct concepts.
In this optional phase, the user may indicate whether they want
Leximancer to automatically identify seed words (and the configuration of
those seed words), or whether the user will manually provide seed words.

Practical: Configuring Concept Seeds Identification

The Concept Seeds Identification settings allow you to choose the number
of concepts (if any) that you would like Leximancer to automatically
extract from the text. To modify these settings, click Edit on the
Concept Seeds Identification in the main interface, and the following
dialogue will appear:

Manual Version 4

74

Automatically

Identify

Concepts

(Yes|No):

Turn

off

automatic

identification of concepts if you would like only concepts that you define
yourself on the map.

Total Number of Concepts (Automatic|Number): This sets the number

of concepts to be extracted automatically. More diverse content requires
more concepts, but less than 100 is recommended. Leaving this setting
on Automatic allows Leximancer to extract the naturally emergent
number of concepts from the data.

Commentary: The larger the data set, or the more conceptually

diverse, the more concepts you should look for. As a rough guide,
think of the number of concepts increasing logarithmically with the
number of documents. However some data, such as magazine
article collections, are very diverse. Note that the selected set of
concepts starts at the top of a ranked list, so you should always get
the most important and general concepts. You need to decide how
deep you want to go into the more specific concepts. Be aware that
if you force more concepts than are really found in the data, you
can start getting junk concepts from among the noise.

Percentage of Name-Like Concepts (Automatic|Number): This setting

allows you to set what proportion of the automatically-extracted
concepts should be forced to be names. Leximancer identifies names by
looking for words that do not begin a sentence but start with a capital
letter.

Commentary: This settings default is Automatic, which creates a

natural mixture of words and names by not forcing any names into
Manual Version 4

75

the list. If you are not interested in names at all, you can set this to
0%. Increase this number if you are particularly interested in
names.

Concept Specificity (On|Off): Concrete concepts differ from normal

concepts in that they are much more specific. They are words that are
strongly related to a small number of concepts, as opposed to normal
concepts that are strongly linked with a large number of other concepts.
Concrete concepts are suitable for creating book indexes, for example.
Because they are so specific, however, you will need many more concrete
concepts to cover the content of a document (which can be very
demanding on resources).

Commentary: Use this with caution. You need lots of memory and
lots of concepts to use this option effectively.

Boilerplate Cutoff (Off-Very Strong): Some concepts may be overly

specific in the sense that they only appear in a particular textual context.
For example, certain words that only appear in repeated addresses,
disclaimers or copyright notices may be identified as a specific concept.
Increasing the Boilerplate cutoff filters out such overly specific concepts
from the automatic list. If this value is set too strong, however, the
remaining concepts may become too general.

Commentary: This feature looks for common words that frequently

occur in the same context and stops them from being selected as
automatic concept seeds. It can be useful for web pages that
contain repeated links or menus, but consider that if set too high,
Manual Version 4

76

it could also cull concepts you might want. Note that this filter does
not remove the words from the text, it just stops them being
selected as automatic concepts. You can still use these words as
manual seeds.

Manual Version 4

77

3. Thesaurus Generation:

You can run this stage of processing using default settings by

clicking the Generate Thesaurus button.

Alternatively you can expand this stage (using the plus sign next to Show
Settings) to Edit the settings for two sub-stages within:

Starting with the seeds automatically extracted by Leximancer, the

Concept Seeds Editing phase (3a) allows users to edit, add or remove
concept seeds from the list.

The following phase, Thesaurus Learning (3b), then generates the

thesaurus of terms associated with each seed. As mentioned earlier,
concepts are collections of correlated words that encompass a central
theme. Once such lists of words have been identified for each concept,
the concept map can be generated to illustrate the relationships between
the concepts in the text.

The learning of the thesaurus associated with each concept is an iterative

process. Seed words are named as such because they start out as being
the central terms of a concept definition - related keywords are collected
during learning. During learning, seeds can also be pushed to the
periphery if more important terms are discovered.

Manual Version 4

78

3a. Practical: Configuring Concept Editing

Clicking on Edit Concept Seeds opens an interface that allows you to edit,
add or delete concepts. This is important for a number of reasons:

automatically extracted maps may contain concepts (such as think

and thought) that are similar, or other concepts that are not of
interest to you.

In the Concept Editing interface you can merge

similar-looking concepts into a single concept, or delete concepts

that you do not wish to see on the map

you may wish to create your own concepts (such as violence) that
you are interested in exploring, or create categories (such as dog)
containing specific instances of terms found in your text (such as
hound and puppy).

Once the Generate Concept Seeds stage has been run, you can check and
modify the discovered concepts by opening the Generate Thesaurus
Settings and clicking on Edit Concept Seeds. The following interface will
appear:

Manual Version 4

79

Here you can edit the concepts extracted automatically by Leximancer in

the Auto Concepts tab, and create your own manual concepts in the User
Defined Concepts tab.
Name-like concepts appear in bold and italicized font in the alphabetical
list.
Prior to thesaurus learning, only the central seed term for each concept
has been identified. In the dialogue above, clicking to select the concept
building reveals a single seed term (identical to the concept name)
comprising the thesaurus for this concept at this stage:

Manual Version 4

80

Single click on concept seeds to select them. Use the Remove button to
remove any unwanted automatic concept seeds. Use the Select All button,
or hold down control <ctrl> while clicking, to select multiple items.
You can merge similar concept seeds by selecting two concepts and
clicking the Merge button. If you do so, the merged concept takes its
name from one of the concept seeds, and the concept then has two
thesaurus items. For example, if you merge the concept seeds company
and companies, the following will result:

Manual Version 4

81

Do not forget to untick the merged item before moving on to work on

other concepts. If you do not untick first, the following action will affect
the merged item as well.

If you change your mind, you can select the merged concept in the list
and use the Unmerge button to separate the two original seeds.

Manual Version 4

82

You can also edit automatically extracted concepts. For instance, if you
wish to add additional thesaurus items, select the concept from the list
and click Edit.
The following dialogue will appear:

This interface allows you to change the name of the concept, identify it as
a name-like or word-like concept, and choose whether to allow
Leximancer to rename the concept during subsequent learning.

You can use the arrow buttons to Add or Remove terms related to the
concept. Add terms if you believe that there are other words that predict
well the presence of the concept in a section of text. You can choose
words from the Frequent Words or Frequent Names lists, or enter your
own words in the Add Words or Add Names text boxes. You can also
identify terms that constitute negative evidence, or evidence that a
concept is absent, but use this option with care. Leximancer will

Manual Version 4

83

automatically learn the weightings for these words from the text during
the Thesaurus Learning phase.
If you wish to create your own concept(s), close this dialogue and return
to the Concept Seed Editor interface. Click on the User Defined Concepts
tab and then click on Add. This opens the Add Concepts interface, where
you can define new concepts yourself.

Name the new concept using the lists of frequent words and names, or
type a concept name into the text box. Use the right arrow button to
move the name of the new concept over to the Current Concepts list:

Click OK to close this window to see your new concept under the Used
Defined Concepts tab in the main Edit Emergent Concept Seeds dialogue.
The new user-defined concept can now be edited in a similar fashion to
automatic concepts.

Manual Version 4

84

If you wish to rerun the project from a prior stage and retain your edits to
emergent concepts, you should click OK to save your edits. Then reopen
the Edit Concept Seeds interface, and Download your edited list of
concepts somewhere on your local drives.

If you run the Generate Concept Seeds stage again, Leximancer will
revert to its original list of auto concepts.

If you wish to use your saved list of edited concepts, go into the Edit
Concept Seeds interface again, and Upload your saved concepts seeds file
in the Auto-concepts tab.

You must save the concept seeds in each of the Auto- and User-defined
tabs separately. The separation affords greater control by allowing you to
reload individual seeds lists if you wish.

Manual Version 4

85

Using Tags
If you have opted to Apply File Tags in the Text Processing settings, then
you should see a tag representing each of your source documents in the
Auto Tags tab in the Edit Concept Seeds interface:

Tag concepts are concepts for which no associated terms will be learned
by Leximancer (unless otherwise instructed). They are useful if you want
to make comparisons among groups within the data.

Manual Version 4

86

You can aggregate the tags using the Merge button, similar to merging
concepts. In this case for example, we could merge the document tags to
create 4 weeks of hearing transcripts for comparison:

If you have created Folder Tags, the numbering lets you know in which
level of the hierarchy a folder resides (Level 1 is the top level).

You can also create User-Defined Tags to perform a simple keyword

search for particular terms of interest.

Manual Version 4

87

The Automatic Sentiment Lens

The Sentiment Lens automatically generates insight into positive

and negative sentiment in your text. A default set of sentiment
concept seeds and their terms are added to your user-defined list.
Sentiment Lens will only apply the sentiment terms that are
identified as relevant and used consistently within your document
set during processing. Sentiment Lens increases both the ease and
accuracy of sentiment analysis
Configuring Sentiment Lens
Under the User Defined Concepts tab, you will notice a button on the
top right: Sentiment Lens. Clicking this button will merge a pre-defined
list of sentiment seeds into your editor:

**Please note: if you make other changes to your user seeds, including
tags, you must make these changes FIRST and then save them by clicking
ok. THEN you may re-enter the Edit Emergent Concept Seeds stage and

Manual Version 4

88

use Sentiment Lens. Leximancer will display a warning dialog if you

attempt to run without saving or discarding changes.
A list of sentiment seeds should appear under the categories _favterms
and _unfavterms. Under the User Defined Tags tab is _negationterms.

- _favterms is a list of commonly used favourable sentiment terms:

approve, best, commend, favour, lauded, praise, great, etc
-

_unfavterms is a list of negative sentiment terms: abhor, anger,

blame, denounce, frown, revolt, etc

- _negationterms is a short list of words causing the following word

to be negated: not, nothing, etc.

Click Okay to return to the control panel.

Manual Version 4

89

If you run the Thesaurus Learning stage, and then return to the User
Defined Concepts tab in the Edit Emergent Concept Seeds stage, you
can observe the effect of Sentiment Lens. Sentiment thesaurus terms that
are irrelevant or inconsistent in your text will be grey. Those left coloured
are suitable for analyzing sentiment in your text and will be used as
thesaurus items to develop sentiment concepts.

Once you reach the Interactive Concept Map, you may also observe new
Sentiment terms that have been automatically added to the Thesaurus.

When analysing news articles about climate change for example, the term
mongering does not appear in the original seed list under _unfavterms.
Yet once Sentiment Lens is applied and run, it appears in the list of
Thesaurus items for _unfavterms (next page):
Manual Version 4

90

Would appear here.

Concept Seed Editor stage

Thesaurus tab in

Concept Map stage

This is because in the climate change literature, the term fear

mongering is used with negative connotations to describe climate
change science. Hence, the automatic Sentiment Lens has picked it up as

Manual Version 4

91

a term that contributes evidence to unfavourable sentiment, even in the

absence of the word fear preceding it.

3b. Generating the Thesaurus

The Thesaurus Learning phase generates a thesaurus of terms associated
with each concept. Concepts are collections of correlated words that
encompass a central theme, for example, the concept client may contain
the terms customer, customers, client, clients, subscriber and
subscribers. The learning of the thesaurus associated with each concept
is an iterative process. Seed words start out as being the central terms of
a concept, collecting related keywords over time. Through learning, the
seed items can be pushed to the periphery if more important terms are
discovered.

Configuring Thesaurus Learning

Clicking on the Thesaurus Settings Edit button reveals the following
interface:

Manual Version 4

92

The Learn Concept Thesaurus From Source Documents (Yes|No)

option allows you to turn off the thesaurus learning and prevents
Leximancer from adding additional items to the concept definitions. This
will result in searches for concepts as keywords, rather than using a
weighted accumulation of evidence. This may be essential for data sets
shorter than a few pages. In such cases, thesaurus abstraction is less
useful due to a smaller vocabulary. Few sensible emergent concepts
would be produced.

Concept Generality (1-21):

This setting allows you to control the

generality of each learned concept. This value is inversely related to the

relevancy threshold required for a word to be included in the concept
thesaurus. Raising this value will increase the fuzziness and generality of
each concept definition by increasing the number of words that will be

Manual Version 4

93

included in each concept. After you have run the learning phase, examine
the log to see how many iterations of thesaurus learning took place to
arrive at the final concept definitions. This number should ideally be
between 3 and 9. If the number is more than 9, consider lowering the
learning threshold. Conversely, if the number of iterations is very low,
consider raising this threshold.

Commentary: This setting controls how easy it is for the thesaurus

learning to grow a concept to a broader set of words. The easier it
is, the more iterations of learning will occur, the more words will be
added to the thesaurus, and the higher the risk of the concept
growing into a more general concept. This can result in the concept
being renamed or being subsumed into another concept which
already exists. If you examine the log file after learning you can
monitor this behaviour. If the data does not consist of natural
language, you should probably disable thesaurus learning, as
described above.

Learn From Tags (Yes|No): You can use this option if you have any tags,
either automatically extracted from tables, file or folder names, or
speaker tags, or manually entered user tags. Turning on Learn From Tags
will treat tags like concept seeds, learning a thesaurus definition for each.
This setting is important if you are conducting Concept Profiling
(discussed below) where you wish to extract concepts that discriminate
between different folders or files (such as extracting what topics
segregate Liberal from Labour party speeches).
Learning Type (Normal|Supervised): There are two forms of learning
that are supported by Leximancer: Automatic and Supervised. Automatic

Manual Version 4

94

is the default behaviour, and in this a concept thesaurus is learned to

characterize a list of seed words. For example, the initial seed word 'dog'
be included in the thesaurus for the Dog concept, which also contains a
collection of other dog-like terms discovered from related text.
Supervised classification, in contrast to Automatic learning, is possible
when you have a complete definition of a category (such as a folder tag,
speaker label, table category, or human coding tag) already embedded in
some training text. In this case, you want to build classifiers that attempt
to faithfully match human classification decisions, rather than discover an
extended thesaurus from seed words. The learned concept should not
include the initial seed item from its thesaurus definition. You are
effectively giving Leximancer examples of a concept that you want
learned. For example, if you were training the system to learn the concept
'violence', you might write the code Violence in locations of the text that
you think are examples of this concept. However, you do not wish the
term Violence to be a crucial term that is required to trigger the
concept. Instead, a concept will be created, encompassing other
discriminating terms from those contexts, that does not include this
supervised term in the extracted classifier.
Sampling (Automatic|1-10): Sampling during learning speeds up the
learning process by only reading every nth block of text. The automatic
setting is normally fine, but you can override this, if necessary, by
choosing n.

Commentary: The automatic sampling setting looks at the size of

the total text data set to decide an appropriate value. The actual
sampling number is increased by 1 for every 15 Mb of text data, so
for anything under 15 Mb, sampling of 1 is used, which means

Manual Version 4

95

every context block is examined for learning. For 15 to 30 Mb of

text data, a sampling of 2 will be used, which means that every
second context block is examined for learning. Note that
classification never uses sampling, and classifies every context
block. Tests have show that classification performance only
decreases marginally if the automatic schedule is followed.

Concept Profiling
These settings allow the learning process to discover new concepts that
are associated with selected user-defined and automatic concepts. This is
useful for profiling concepts or names, for doing discriminant analysis on
prior concepts, or for adding a layer of more specific concepts which
expand upon a top layer of general concepts. Profiling also allows you to
ignore large sections of text that are not relevant to your particular
interests.

Once the initial concept definitions have been created, words that are
highly relevant to these concepts can be identified as potential seeds for
new concepts. For example, if you profile the initial seed flowers, a
concept definition is grown around this word as usual. Then new
concepts are developed from the flowers definition that would produce
more specific topics, such as roses, daffodils, petals, and bees.

This process is useful if you are trying to generate concepts that will
allow segregation between various document categories. For example, if
you are trying to discover differences between good and bad applicants,
simply place exemplars of each in two separate folders (one for each
type), and Apply Folder Tags in the Preprocessing stage. This will create a
Manual Version 4

96

concept class for each folder. In the Concept Editor, only retain these
folder tags in your Automatic Concepts and Tags lists. Switch on Learn
From Tags in the Thesaurus Learning phase, and use the profiling
settings described below to extract relevant segregating concepts.

Number to Discover (Off|10-1000): This parameter specifies how many

concepts should be profiled or discovered from the pre-defined
concepts. More pre-defined concepts normally require more discovered
concepts, but less than 100 is recommended. As a guide, select between
3 and 10 discovered concepts per pre-defined concept to give a
reasonable coverage.

Themed Discovery (Concepts in ALL| Concepts in ANY| Concepts in

EACH): Choose how you want the discovered concepts to be related to
the pre-defined concept set: ANY gives the Union (related to concept1 OR
concept2 ), EACH gives the Exclusive Disjunction (XOR: related to
concept1 OR concept2 but not both), and ALL gives the INTERSECTION
(related to concept1 AND concept2). Choosing the intersection of the
concepts will only extract concept seeds that are highly relevant to all or
most of the learned concepts. For example, conducting a themed
discovery from the concepts sun, surf, and sand may lead to concepts
more relevant to beach scenarios than using words relevant to only one
of these concepts (ie: the union of the concepts). XOR is designed for
strong discrimination of target classifications (ie: finding specific
concepts that segregate between the predefined concepts).

Commentary: If the pre-defined concepts surround some theme,

such as say beach life or environmental issues, you probably want
the discovered concepts to follow the theme, so choose the AND
Manual Version 4

97

(intersection) option. If you want to discover concepts that strongly

discriminate between the pre-defined concepts or tags, choose the
XOR (disjunction) operator.

Only Discover Names (Yes|No): This option lets you discover name
concepts only when profiling. This is useful for discovering social
networks of association.

Manual Version 4

98

4. Run Project:

You can run this stage of processing (and all of these preceding it)
using default settings by clicking the Run Project button.

Alternatively you can expand this stage (using the plus sign next to Show
Settings) to Edit the settings for three sub-stages within:

4a. Editing Compound Concepts

Manually compound selected concepts via Boolean operators to obtain
deeper and more meaningful analysis.
Clicking Edit at the Compound Concepts stage opens this interface:

Manual Version 4

99

Select the boxes for any concepts you wish to combine into compound
concepts. There are lists of all tags, name concepts and word concepts in
the left and centre columns. Move them across to the compounding
workbench using the right-hand arrow:

Manual Version 4

100

Combing the concepts using the AND operator requires both concepts to
appear in the same (2-sentence) piece of text for the compound to be
coded.

After moving the concepts you wish to combine into the right column,
tick both of them again and click the AND button in the header:

Manual Version 4

101

Hover your mouse over the compound concept to see the equation
defining it.

Note: Concepts used to build a compound remain available as singular

concepts in the project results as well (unless you exclude the constituent
concepts from the Mapping Concepts list in the Concept Coding
Settings).

Use the same method to combine concepts with the Boolean operator
OR. Using the OR conjunction means that evidence for your compound
concept will be calculated to include evidence for either of your concepts.
Compounding concepts using the OR operator is similar to Merging
concepts in the Edit Concept Seeds interface.

Manual Version 4

102

To make a compound concept using the NOT operator, first select the
concept you wish to negate. The NOT button at the top of the column
will become available. Clicking the NOT button will negate the concept
you have selected:

You can now combine the negated concept with another concept by
following the steps for combining concepts with AND as outlined above.
Doing so will include instances of the positive concept, and exclude
instances of the negated concept:

Manual Version 4

103

The same procedures outlined above can also be used to build more
complex Sentiment concepts.
This is done automatically when you click Sentiment Lens in the Concept
Seeds Editing interface. For example, the Sentiment Lens creates a
compound concept for Positive Sentiment that includes favourable terms
and excludes negation terms:

Manual Version 4

104

You can specify different rules for coding sentiment by combining the
concepts used as building blocks for the Sentiment Lens (favterms,
unfavterms, negationterms) in different ways if you wish.

Finally, once you have created a compound concept you may edit it. Do
so by selecting a compound, and then clicking Edit:

Manual Version 4

105

This allows you to rename your new compound concept, specify whether
it is a name or a regular word, and whether or not it should be treated as
sentiment:

(NOTE: Sentiment terms will not appear on the map, but will be present in
the report tabs to the right of the map).

Manual Version 4

106

Compound concepts will automatically appear on the map if they are

regular word concepts. Name-like compounds need to be specifically
added to the Mapping Concepts list in the Concept Coding Settings.

After Clicking Ok to leave the Compound Concepts interface, click the

Code Concepts Settings Edit button to open this interface:

Select the compound concept you wish to add to the map, and move it to
the left hand column using the arrow.

Now when you click Ok, Run the Project, then open the concept map,
your newly created compound concept will appear:

Manual Version 4

107

Manual Version 4

108

4b. Concept Coding

These are the concepts that appear on the conceptual map, and represent
the top-level of classification of the text. Generally all concepts can be
used as mapping concepts, but there are some cases in which you may
want to map only a subset of the concepts. For example, you could map
only the concepts discovered automatically by Leximancer in once
instance, and then only the names seeded by you the user in another.
You can also choose to map specific names and concepts from the lists
provided if you wish.

Configuring the Code Concepts Stage

The first part of configuring the Concept Coding phase is to specify which
of the Available Names and Concepts you wish to include on your map.
Clicking on Edit for the Concept Coding phase opens the following
interface:

Manual Version 4

109

By default, All Concepts and All Discovered Names appear in the Mapping
Concepts tab. This means that all word-like and name-like concepts
discovered by Leximancer will appear on the concept map. Tags,
compound concepts and name-like user-defined concepts must be
added to the list manually. Using the arrows to replace these wildcards
with others from the General list allows you map other groups of
concepts:

Discovered names and concepts refer to those automatically-identified by

the program, and User names and concepts refer to those created by you
the user. Tags are treated as User-defined.

Instead of using the option in the General list, you can choose to map
particular names (including tags) and concepts using the lists on the
right. Simply use the arrow buttons to add or delete concepts from each
of the lists.

Manual Version 4

110

This interface also allows you to filter records in and out of the analysis
by specifying Kill Concepts and Required Concepts. Simply click on the
appropriate tab and move the desired concept(s) or tag(s) across using
the arrow buttons.

Kill Concepts and Required Concepts

Kill concepts are concepts that if found in a classified block of text,
cause all other classifications of that block to be suppressed. For
example, you could use this option to suppress the processing of
questions asked by an interviewer to focus on the responses of the
interviewee. If you identified the dialogue spoken by the interviewer using
the correct form (speaker name starting with a capital letter on a new
line, and followed with a colon and a space, eg: Alan: ), Leximancer has a
setting in the Preprocessing Options called Apply Dialogue Tags which
will automatically identify the dialogue tags. These will be presented to
you in the Auto Tags tab in the Concept Seed Editor. You can then set
the interviewer dialogue tag as a Kill class to suppress the processing
text spoken by the interviewer.

Required concepts by contrast, are classifications that must be found in

blocks of text, or else the blocks are ignored. That is, at least one of the
required classifications must be found in any context block for it to be
included in the concept map.

Options (Classification Settings)

Once the concept definitions have been learned, each block of text is
tagged with the names of the concepts that it contains. This process is

Manual Version 4

111

similar to manual coding used in content analysis. However, the benefit

of using Leximancer for this task is that it is fast (compared to the time
taken for human coders), and is more objective (as opposed to humans,
where there is much variability in coding performance).

In Classifying the text document, the following steps are taken:

The text is broken into context blocks of n sentences

For name-like concepts, all the associated terms that are present in
the block are noted. The block is said to contain the concept if the
word with the highest relevancy to the concept is above a set
threshold.

For word-like concepts, the relevancies of all the associated

keywords that are present in the block are summed. The block is
said to contain the concept if this sum is greater than a predefined
threshold.

Classification Settings
Clicking on the Options tab opens the following dialogue:

Word Classification Threshold (2-4.9): For word-like concepts, the

relevancies of all the associated keywords that are present in the block
Manual Version 4

112

are summed. The block is said to contain the concept if this sum is
greater than a predefined threshold. This threshold specifies how much
cumulative evidence per sentence is needed for a word concept
classification to be assigned to a context block.

Commentary: Note that the actual threshold used is this value

multiplied by the average number of sentences per context block
found in the data, not the number of sentences found in any
particular segment of text. This means that a fixed threshold is
applied for all context blocks. It may seem that the actual threshold
should be calculated from the number of sentences found, but this
would mean that less evidence would be required to trigger a
classification in some places than others. After all, one sentence is
less evidence than three. The units of this threshold are the same
as the relevancy standard deviation values shown for terms in the
thesaurus, so you can get a feeling for how much cumulative
evidence you need by looking at the learned thesaurus which can
be viewed through the map interface.

Name Classification Threshold (2.6-5): For name-like concepts, all the

associated terms that are present in the block are noted. The block is
said to contain the concept if the word with the highest relevancy to the
concept is above the threshold value specified here. That is, this value
gives the minimum strength of the maximally weighted piece of evidence
to trigger a name classification.

Commentary: The idea behind this threshold is that one strong

piece of evidence is enough to indicate the presence of a named
thing such as company. For example, the name of the CEO or the
Manual Version 4

113

stock exchange code would be enough to indicate the company is

involved. However, lots of weak evidence for the company is not
sufficient, as it could be describing the same area of the market but
actually a competitor company. This is related to the notion that a
named thing is an instance of a class, rather than a class. If you
want cumulative tagging of named things based on a similar lexical
context, use the Treat Names as Words option. Like the Word
classification Threshold, the units of this threshold are relevancy
standard deviation units as shown in the thesaurus.

The next step in processing is to measure the co-occurrence of the

identified concepts in the text (indexing) for the generation of the
conceptual map. The positions of the discovered concepts and groups of
important co-occurring concepts are also noted for use with the query
browser.

Classification

Threshold

for

Supervised

Classifiers

(0.7-1.4):

Supervised Concepts are used to find instances of a particular concept in

the text. Generally, such concepts are 'trained' by tagging exemplars with
a code word (such as 'violence'). They are trained to be sensitive to the
vocabulary surrounding such tags, without including the tag itself within
the definition. Thus, generally there is weaker evidence compared to
normal concepts such as 'dog' in which the key word is present in the
script. For this reason, supervised concepts require a lower classification
threshold. This threshold specifies how much cumulative evidence per
sentence is needed for a supervised classification to be assigned to a
context block.

Manual Version 4

114

Treat Name-Like Concepts as Word-Like (Yes|No): This setting forces

name-like concepts to be classified using the same system as word-like
concepts, allowing more intuitive coding. This option allows tagging of
named things based on similar lexical context, rather that similar
identity.

Kill Concept Scope (Kill Whole Document|Only Kill Context Block):

This option lets you choose whether to suppress the classification of the
entire document should a killed class be present, or just the context
block in which the kill class is located.

Generate Table of Text Segment Classifications (Yes|No): If this option

is enabled, a delimited text output file listing the concepts tagged in each
section of text through the data file is created. This output file, called
Table of Text Segment Classifications, is accessible under the Export tab
once processing is complete.

Statistics Type (Count|Weight): This setting affects the type of statistics

produced in the High Resolution Data Output. The output file lists the
concepts tagged in each section of text through the data. If the Generate
Table of Text Segment Classifications option is enabled, this file is
created and accessible under the Export tab. Normally an assigned tag is
simply counted. You can change this setting so that tags are assigned a
confidence weight. This results in weighted sums rather than count
statistics.

Generate Document Section Classification (No Output|Document as

Vector|Document as Matrix): Some advanced applications require
classification metadata for each document.
Manual Version 4

115

If Document Metadata is set to Document as Vector, an XML file is created

that lists the concepts tagged in each document or document section.
This output enables analysis of the conceptual content of the data by
document section.

If Document Metadata is set to Document as Matrix, an XML file is

produced that reports the matrix of concept co-occurrences within each
document or document section. Matrix classification enables more
accurate document comparison.

If either of these outputs is created, it is called Document Section

Classification, and is accessible under the Data Exports button below the
main Project Control Panel, or via the Exports tab in the concept map
interface.

4c. Project Outputs

The last settings that can be edited through the main interface is Project
Outputs. In this phase, the map displaying the relationship between
variables is constructed. Clicking Edit at the Project Output Settings
stage opens the following interface:

Manual Version 4

116

Generating the Concept Map

One of the principal aims of Leximancer is to quantify the relationships
between concepts (i.e. the co-occurrence of concepts), and to represent
this information in a useful manner (in a concept map) that can be used
for exploring the content of the documents. The concept map can be
thought of as a birds eye view of the data, illustrating the main features
(i.e. concepts) and how they interrelate.

The mapping phase generates a two dimensional projection of the

original high dimensional co-occurrence matrix between the concepts. It
must be emphasised that the process of generating this map is
stochastic. Concepts on the map may settle in different positions with
each generation of a new map.
In understanding this, consider that concepts are initially scattered
randomly throughout the map space. If you imagine the space of possible
map arrangements as a hilly table top, and you throw a marble from a

Manual Version 4

117

random place on the edge, the marble could settle in different valleys
depending

on

where

it

starts.

There

may

be

multiple

shallow

valleys (local minima) in the map terrain if words are used ambiguously
and the data is semantically confused. In this case the data should not
form a stable pattern anyway. Another possibility is that some concepts
in the data should in fact be stop words, but aren't in the list. An example
of this is the emergence of the concept 'think' in interview transcripts.
This concept is often bleached of semantic meaning and used by
convention only. The technical result of the presence of highly-connected
and indiscriminate concept nodes is that the map to loses differentiation
and stability. The over-connected concept resembles a mountain which
negates the existence of all the valleys in the terrain. To fix this, remove
the over-connected concept.

The practical implication is that for a strict interpretation of the cluster

map, the clustering should be run several times from scratch and the
map inspected on each occasion. If the relative positioning of the
concepts is similar between runs, then the cluster map is likely to be
representative.

Note

that

rotations

and

reflections

are

permitted

variations. If the map changes in gross structure, then revision of some of

the parameters is required. The concept map display supplied with
Leximancer allows easy animated re-clustering and map comparison
using the buttons above the map.

Topical versus Social Mapping

In concept mapping, there is one setting to choose, namely whether to
use a Topical or Social map. The Social map has a more circular
symmetry and emphasises the similarity between the conceptual contexts
in which the words appear. A Social map is best when entities tend to be
Manual Version 4

118

related to fewer other entities, such as a map made up of many name

concepts.
The Topical map, by comparison, is more spread out, emphasising the
co-occurrence between items. It tends to emphasise differences and
direct relationships, and is best for discriminant analysis. The Topical
map is also much more stable for highly connected entities, such as
topics. The most common reason for cluster instability is that the
concepts on the map are too highly connected, and no strong pattern can
be found. The Topical variant of the clustering algorithm produces more
stability in maps of this kind, so switching to this setting will often
stabilise the map. However, the most important settings which govern the
connectedness of the map are the classification thresholds and the size
of the coded context block, which are located in the Classification
Settings in the Locate Concepts node. If the coded context block is too
large, or the classification threshold is too low, then each concept will
tend to be related to every other concept. If you have some highlyconnected concepts which are effectively bleached of meaning in your
data, removing from the concept lists in the Concept Seeds Editor will
often stabilise the map. Words such as sort, think, and kind often
appear in spoken transcripts and may be used as filler words which are
essentially stopwords. Inspect the actual text locations to check the way
words like these are being used before removing them.

In summary, the Topical clustering algorithm is more stable than the

Social, but will discover fewer indirect relationships. The cluster map
should be considered as indicative and should be used for generating
hypotheses for confirmation in the text data. It is not a quantitative
statement of fact.

Manual Version 4

119

Default Theme Size Percentage (10-65): This option sets the size of
themes (concept groupings) visible when the map interface opens
initially. The theme size slider beneath the concept map also allows this
parameter to be adjusted through the map interface.

Create Theme Hierarchy (Yes|No): If this option is enabled, an xml file

is created that contains a nested hierarchy of the discovered concepts.
The hierarchy is induced from the discovered hierarchy of concepts
shown in the Leximancer concept map. This output file, called Table of
Text Segment Classifications, is accessible under the Export tab.

You can also specify the Size (Width and Height) of the concept map to be
produced in this interface.

Final Outputs
The Concept Map

Manual Version 4

120

See page 8 for an explanation of the Concept Map.

The Concept Cloud

Manual Version 4

121

See page 15 for an explanation of the Concept Cloud.

Manual Version 4

122

The Insight Dashboard Report

The Insight Dashboard is a standalone pdf or html report that can be
produced within a Leximancer project. The Dashboard report adopts a
more quantitative focus than the concept map, and is designed to provide
a quick understanding of the project results.

The report is not set up to be created automatically when you run a

project, as not all projects are suitable for the dashboard report format.
The Dashboard is designed for some sort of comparison, or difference
analysis. Usually you must have created some (source document, tabular,
speaker or folder) tags to use as categories for comparison in the
Dashboard report. You can then use the Report to investigate the
Attributes (independent variables) associated with certain Categories
(dependent variables) in the data.
In this example, File tags were aggregated to create a category for each
week of hearing transcripts after an explosion at a plastics factory. The
Dashboard quadrant below shows the concepts most relevant to the
discussion in each week (trends). The concepts towards the upper, righthand (magic) quadrant were talked about most frequently in each week,
and are most characteristic of the discourse during that time:

Manual Version 4

123

The Dashboard is produced as a standalone document so that it can be

shared with others. Once created, it can be Downloaded from the Project
Control header, or from the Data Exports beneath the main Project
Control Panel. It can then be printed or sent to colleagues, collaborators
and clients. The report does not require any knowledge of Leximancer to
be understood.

Manual Version 4

124

The Dashboard is intended as a general purpose reporting tool, and the

user can configure the options to suit their particular application or
research question.

Configuring the Insight Dashboard Report

To configure a Dashboard Report, expand the Run Project Settings, and
Edit the Project Output Settings. Click on the Insight Dashboard tab to
reveal the following interface:

Tick the Generate Insight Dashboard option.

Tick the Generate Quadrant Report option. This includes a 'magic

quadrant' graphic in the Report for identification of key issues.

Manual Version 4

125

Set the Statistics Type setting (Segment|Segment)

Using the Segment Statistics setting, concepts are coded and classified at
the level of the text segment. You can define a text segment (the coding
resolution) using the Sentences per Block setting in the Pre-processing
Options.

Using the Section Statistics setting, concept codes are applied to sections
of the data. The definition of a document section depends on the type
of data:

If you are processing Microsoft Word or pdf documents, then a

document section is usually an individual file.

If you are processing delimited spreadsheet data, however, then a

section is a single free text cell. This setting can be used to report
the number of responses (whole comments) coded with particular
concepts.

The Truncate Ranked Concepts setting allows you to specify the

number of Attributes reported for each Category. You can control
the level of detail in the Report using this setting. The Do Not
Truncate default does not apply any cut-off, but presents all the
Attributes associated with each Category.

The Auto Scale setting scales the axes in the Quadrant graphic.
This spreads the Attributes in the Quadrant space for improved
visibility.

Manual Version 4

126

If this option is enabled, the Prominence statistics (described in the

preamble to the Report) are expressed as relative probabilities.

The Include All Text Segments setting introduces an optional final

section to the Report that presents all text segments matching
each concept query.

Specify the Categories (or dependent variables) of interest. You can

specify any number of Categories, though using upwards of 10 can
clutter the Quadrant graphic.

Often the Categories will be (auto- or user-defined) Tags identifying

different levels of variables, or groups for comparison, within the data.
Gender (male or female) and tone (favourable or unfavourable) are
examples of possible Categories.

Select the Tags (or concepts) of interest from the Available Names
(or Concepts) list(s), and use the appropriate left arrow to add
these to the Categories list.

Note: If you wish to use Tags as Categories, you must add these to the
Mapping Concepts list by hand in the Select Concept to Locate phase.
This codes the data with the Tags so that they can be used in the
Dashboard Report. It will also cause them to clustered on the map among
the topical concepts.

Specify some Attributes (or independent variables) of interest.

Manual Version 4

127

You might use the emergent concepts as Attributes in the Dashboard

Report. In this case, the Report will compare the concepts associated with
each of your Categories

Select the All Concepts wild card from the General list, then click
the Attributes left arrow to use all the word-like concepts as
Attributes. This wildcard includes all the entries in the Available
Concepts list on the right. Alternatively, you can select individual
concepts from the list and add them as Attributes by hand.

Select the All Discovered Names wild card from the General list,
then click the Attributes left arrow to use all the word-like concepts
as Attributes. This wildcard includes all the entries in the Available
Names list in the centre. Alternatively, you can select individual
names from the list and add them as Attributes by hand.

When the settings have been configured, click Ok and run the final
stages of processing to produce the Dashboard Report.

After clicking the Run Project button, you can download the Dashboard
Report from the button beneath the main Project Control Panel.

The Dashboard can be downloaded in pdf of html format.

The pdf version can be viewed in Adobe Acrobat Reader. The

Quadrant graphic requires at least Acrobat version 9 to be
displayed.

Manual Version 4

128

The html version is useful if you wish to make edits to the Report.
This version can be saved as a zipped folder (or archive) on your
local machine. You may need to rename the folder (changing the
extension from insight-dashboard-zip to insight-dashboard.zip) to
allow it to be extracted or opened. Click on the insightdashboard.html file to view the Report in a browser tab, or right
click and select Open With to view the Report in another application
(such as Microsoft Word).

The Dashboard is named after the project in which it was created. The
header provides counts of the Total text Segments or Sections coded in
the Report. It also presents counts for the number of Concepts and
Categories specified.

Note: The preamble explains the various sections of the Report, and
is included as part of Dashboard to allow others to understand its
contents.

The Dashboard is designed to be interactive, and the Table of Contents

includes clickable links, as do most many other sections of the Report. In
the html version, you may have to hold down the <control> key and click
to navigate the Report links.

Explanation of the Statistics in the Insight Dashboard Report

The Frequency axis on the Quadrant graphic represents a conditional
probability. Given that a text extract comes from a particular Category, it
gives the chance that the Attribute is coded in this text extract. This
measures frequency of mention in the data, and is affected by the
Manual Version 4

129

distribution of comments across the Categories. The frequency score is in

fact a log scale (so that it can be mapped on the quadrant).

The Strength score is the reciprocal conditional probability. Given that the
Attribute is present in a section of text, it gives the probability that this
text comes from that Category. Strong concepts distinguish the Category
from others, whether or not the Attribute is mentioned frequently.

The percentages in the Ranked Concept for Categories lists match the
quadrant coordinates. They reflect the same Strength and Frequency
conditional probabilities.

The Prominence score combines the Strength and Frequency scores using
Bayesian statistics. Prominence scores are absolute measures of
correlation between category and attribute, and can be used to make
comparisons over time.

Manual Version 4

130

Data Exports
Among the projects results, Leximancer produces several statistical
reports. These can be exported for reporting or to allow further analysis
in other applications.

Several reports are available for Download from the Data Exports button
beneath the Project Control Panel. These include: (1) the Pairs of
Concepts across the Entire Corpus (the co-occurrence matrix); (2) the List
or Pairs of Concepts within each Text Excerpt; (3) the List of Concepts
within each Context Block, within each Text Excerpt; and (4) The
Sentiment Lens Seeds Set:

Manual Version 4

131

Hover your mouse over the Download button for a description of the level
of detail in the each of the reports.

Leximancer Exports are formulated from the base Data Corpus and
loaded into the system for analysis. The granularity of the Exports is
denoted by:

a Text Excerpt, being defined as:

o

a document - if working with a document set, or

the contents of a text cell - if working with spreadsheet data

a Context Block, being defined as:

o

the number of sentences that forms the context 'boundary'

or 'window' for automatic Concept determination, as defined
within Leximancer in the Pre-Process editing stage

There will typically be a number of Context Blocks within a Text

Excerpt.
Available exports include:

1. Pairs of Concepts across the entire Data Corpus (the cooccurrence matrix)
Downloads a comma delimited file displaying the matrix of cooccurrences between concepts. This file will open a spreadsheet program,
including recent versions of Excel. It contains co-occurrence counts,
listed for every concept pair combination, as well as x,y coordinates for
each concept on the map, and the weight for each concept, which is the
sum of its co-occurrences with all the other concepts:

Manual Version 4

132

2. List or pairs of Concepts within each Text Excerpt

This CSV or XML file contains classifications of text excerpts. Each block
(row) indicates:
- the file and section number for the text excerpt;
- the surrogate id for viewing this context block;
- the tabular input data, a field for each quantitative column in the
input data;
- for tabular input, a field whose value indicates which text column in
the row is indicated. This is for situations where there are multiple
text cells in a row of input data;
- a nested block containing classifications assigned to this text
excerpt, with occurrence counts and cumulative weights

This output is designed for import of document or text excerpt

classifications into a database. This output file is not sparse.

3. List of Concepts within each Context Block, within each Text

Excerpt

Manual Version 4

133

This tab delimited text file contains one row for each context block. Each
row indicates:

- the file, text excerpt, and starting sequence number of the context
block;
- the html surrogate link for viewing this context block in a browser;
- the presence or absence of a concept or tag in the context block.

There is one column for each concept or tag class. As a result, this table
is very sparse.

This import is specifically designed for input into statistics or data mining
packages for building models such as: decision trees, rule sets, logistic
regression, or market basket analysis. There is a setting in the
Classifications Settings tab to generate either real valued or binary values.

4. Sentiment Lens Seed Set

This export will open a list of all the Sentiment seeds remaining after
they have gone through the tuning process of Sentiment Lens.

Manual Version 4

134

This file can be Uploaded directly into other projects via the Concept
Seeds Settings Edit interface.

Logbook Exports
In the map interface, Leximancer allows complex records of queries to be
stored and exported. Find a particular query of interest, and its example
text, and add it to the logbook:

Manual Version 4

135

Go to the Logbook tab to review all the logged text:

Manual Version 4

136

From here, you can export either just the current page, or every entry in
the logbook:

Manual Version 4

137

Your exported logbook entries will pop up in a new window of your

browser. Pop-ups need to be enabled for this to occur. Leximancer will
have logged the Document ID, Folder/s, Author, Date and Time, Query
Terms, and the example text:

Section 5: Example Advanced Techniques

Manual Version 4

138

This section is intended to be a guide for some of the styles of analysis

that are possible with the Leximancer system. However, the system is
designed to be a general-purpose tool, and the confident user is
encouraged to experiment.
Specific tutorials include: Manual Concept Seeding; Profiling; Profiling
Using Tag Categories; Extracting a Social Network; Analysing Transcripts;
Analysing Spreadsheet Data

1. Manual Concept Seeding

You can seed your own concepts in the User-Defined Concepts tab if you
Show the Generate Concepts Settings, and click Edit Concept Seeds
Settings. Move to the User-Defined Concepts tab. If you create new
manual seeds, thesaurus definitions will be extracted for these and any
automatically-identified concepts during the Generate Thesaurus phase.
You can also create your own Manual Tags. These act like lists of
keywords, or fixed dictionaries they are not modified by the learning
process.

In many cases, an automatically generated map may contain concepts

that are irrelevant to your interests or concepts that are similar (such as
thought and think), or the map may be lacking concepts that you wish to
locate in the text. You can seed your own concepts prior to running the
Thesaurus Learning phase by clicking on Edit Concept Seeds in the
Project Control Panel. There you can add, edit, merge or delete concepts
in order to produce cleaner or more tailored concept maps. If you add
(User-Defined) concept seeds manually, thesaurus definitions for these
will be extracted from the text along with any automatically-extracted
concept seeds. You can also create your own User-Defined Tags and

Manual Version 4

139

these act like lists of keywords, or fixed dictionaries they are not
modified by the learning process.

Configuring Manual Concept Seeding

You can Add, Edit, Remove, Merge or Unmerge concepts in Generate
Thesaurus, then Edit Concept Seeds. This is important for a number of
reasons:

Among the Auto Concepts (those automatically-extracted by the

software), there may be concepts seeds (such as think and thought)
that are similar, or seeds that are not of interest to you. You can
Merge similar concept seeds into a single concept, or Remove
concepts that you do not wish to see on the map

in the User Defined Concepts tab, you may wish to Add your own
seeds (such as violence) to search for concepts that you are interested
in exploring, or create categories (such as dogs) containing the
specific instances found in your text (such as hound and puppy).

In order to modify concepts automatically extracted by Leximancer,

Generate Concept Seeds (the node prior to Generate Thesaurus) needs to
have been run. If this node has been run (i.e. if you have previously
generated a map), it will be green and its status will say Ready. If not,
click on the Generate Thesaurus button to run this stage.
After Generate Concepts has been run, showing the Generate Concepts
Settings, and then clicking on Edit Concept Seeds Settings reveals the
following interface:

Manual Version 4

140

There are tabs for editing the Automatic Concepts (concepts identified by
Leximancer) and User Defined concepts (concepts that you wish to define
yourself).
At this stage, only the central key word for each concept has been
identified. The learning of associated terms and their weightings occurs
in the following Thesaurus Learning phase.
In the interface above, you can select and Merge or Delete concept seeds
(note: holding down <ctrl> while clicking allows you to select multiple
items).
Open the User-Defined Concepts tab and click on Add to open the Add
Concepts interface:

Manual Version 4

141

You can select terms from the lists of Frequently-occurring Words or

Names on the left, or type in your own name for the new concept using
the text boxes beneath the lists. Use the arrow buttons to move the term
to the right to create your concept. Click on Ok to return to the Concept
Seed Editing interface.
To add thesaurus terms to the concepts early definition, select the
concept in the Auto- or User-Defined concepts tab, and click on Edit. The
following interface will open:

Manual Version 4

142

Here you can add terms strongly-related to your concept. For example, if
you are interested in finding sections in your text containing violence,
create the concept violence and add any terms from the Frequent Words
or Names list above that you think indicate a violent act. Only use words
that fairly unambiguously indicate this concept in the text. Leximancer
will automatically find additional terms from the text during the
Thesaurus Learning phase, so you don't have to know all the relevant
words in advance. Click on Ok to save your changes and exit.
When you return to the Concept Seed Editing interface, you can also
create and edit Tag categories. These are concepts for which no
associated terms will be learned by Leximancer. This useful if you want to
compare groups in the data (using file or folder tags) or perform a simple
keyword search for terms.
Click on OK to close the Concept Seed Editor and return to the main
Project Control Panel. Click on Run Project to run the remaining phases

Manual Version 4

143

on default settings. Click on Concept Map to view the map containing the
new concepts.

2. Profiling
This function is not the same as automatic concept discovery. The aim
here is to discover new concepts during learning which are relevant to the
concepts defined in advance, either in the Automatic- or in the UserDefined Concepts tabs. For example, this setting would allow you to
extract the main concepts related to stem cell research from a broader
set of documents. Concept profiling settings can be found under Edit for
the Thesaurus Settings in the Generate Thesaurus stage. Note that Tags
do not take part in this process automatically. If you have Tag categories,
folder tags for example, which you wish to profile, you must turn on the
Learn From Tags option in the Thesaurus Settings. It is important to
understand that although these discovered concepts are seeded from
words that are relevant to the prior concepts, they are then learned as
fully-fledged independent concepts. As a result, the map will contain
some peripheral areas of meaning that do not directly contain the prior
concepts. Contrast this with the Required Concepts function described
below.

The profiling function has three alternative behaviours: ALL, ANY, and
EACH. You can ask for the related concepts to be relevant to many of the
prior concepts, and thus follow a theme encompassed by all the prior
concepts this is the ALL option, and resembles set intersection.
Alternatively, the discovered concepts need only be related to at least one
of the prior concepts this is the ANY option, which is similar to set
union. The EACH option discovers equal fractions of profile concepts for
each predefined concept, and these concepts show what is peculiar to

Manual Version 4

144

each predefined concept. The EACH option is very useful for enhanced
discrimination of prior concepts.
If you wanted to extract the main concepts related to stem cell research
from a broader set of documents, for example, you could disable
Automatic Concept Identification in Concept Seeds Edit in the Generate
Concepts stage. Instead you would create user-defined seeds for multiple
simple concepts that encompass the scenario. You might seed concepts
such as research, ethics, debate and so on. Keep the seed words for
each concept simple, and dont try too hard to restrict the seeds of each
concept to just the topic you are after. In this instance we will be
considering the intersection of all these elements. Expand the Generate
Thesaurus Settings, and Click on Edit for Thesaurus Settings. Specify a
quota of concepts to be discovered in the Concept Profiling options.
Choose to discover several profiled concepts per prior concept. Select
Concepts in ALL as the operator.
If you are attempting to discover the network of associated names around
a name or a scenario, you can choose to only discover name-like
concepts during profiling. You should try the Social mapping algorithm
first for this style of map.

It is important to understand that although the profiled concepts are

seeded from words that are relevant to the prior concepts, they are then
learned as fully-fledged, independent concepts. As a result, the map will
contain some peripheral areas of meaning that do not directly contain the
prior concepts. Contrast this with the Required Concepts function
described below.

Manual Version 4

145

Configuring Concept Profiling

Leximancer will generally extract the main concepts that occur within
your documents. However, in some instances you may be interested in
inspecting certain aspects of your text in more detail. For example, given
1000 newspaper articles, you may only be interested in the events
containing references to violence. In such cases, you can use profiling to
extract concepts that are specific to violence, rather than those that occur
in all the articles that you have been given.
To profile an issue:

Select your data files as usual, expand the Generate Concepts

Settings.

Click on Edit Concept Seeds Settings to untick the Automatically

Identify Concepts box, because you don't want any concepts
present other than the ones you are interested in profiling.

Run the Generate Concepts stage.

Expand the Generate Thesaurus Settings, and click on Edit Concept

Seeds (as discussed above- Configuring Concept Editing) to create
one or more concepts to profile. In this case, a concept called
pipework is of interest for profiling.

Then click on Thesaurus Settings, and enter the number of

concepts that you would like to see on the map related to this
theme (60 in this example) in the Concept Profiling section:

Manual Version 4

146

Click OK to exit the Thesaurus Settings. Click on Run Project in the

main Project Control Panel to run all remaining stages of
processing, then click Concept Map to reveal the profile map.

The concept map will contain profiled concepts specific to your area of
interest. The map below is a profile of the Pipework concept in the
Explosion Enquiry (hearing transcript) data set:

Manual Version 4

147

Concept profiling is very useful in cases where you have a large amount
of text, but are only interested in exploring particular themes.
Profiling also allows you to make conceptual comparisons (highlighting
divergence or consensus) between groups in your data. This strategy
combines the use of tag categories and profiling, and is described in the
next section (Profiling using Tag Categories).

3. Profiling Using Tag Categories

Manual Version 4

148

Folder and Filename Tag Generation

Filename or folder tagging converts categories of text (or variables) into
explicit tags in the text, and hence into concepts on the map. The names
of all parent folders of a file, and optionally the name of the file itself, are
embedded as separate tags on each sentence within a file. For example, a
file called patient1 inside the folder control1 below the data folder
would have separate tags [control1] and [patient1] inserted in each
sentence (if folder and filename tags are applied). These tags are
imported into the Automatic Tags Tab so long as Automatic Concept
Identification is enabled. Note that the hierarchical order of the folders is
not important, and is not carried into the Leximancer map. If you have a
folder called Control under several different branches, then this
becomes one concept called Control on the map, and this can be a
powerful feature for freeing up the exploration of category co-variances.

Comparing Names or Categories based on Shared Semantics

The strategy here is to compare categories based on shared concept
usage. The categories are usually Tags, often generated using the folder
tags setting. You might have several submissions on an issue from
different people and organizations, and you want to know how they focus
on the major topics. Enabled the Apply Folder Tags setting will cause
each file to generate a tag. This is also very useful for looking at trends
over time. Simply put documents into folders by month or by year etc.
Use automatic concept identification to generate a set of concepts that
characterise the whole text collection, then complete the map, making
sure that you add the tags of interest to the Mapping Concepts list in the
Project Output Settings in the Run Project stage. Make sure that you are
using the Topical network algorithm to cluster the concept map. The map
Manual Version 4

149

should then show the tag categories distributed around the concepts. If
the locations of the tags relative to the concept field show little
repeatability, then you should conclude that the tag categories are
difficult to differentiate based on the global concept selection. Essentially,
they all address most of the same global concepts to similar degrees.
This is a result in itself, but if you actually wish to discriminate between
tag categories, see the section on Discrimination of Concepts, Names, or
Categories based on Semantics.

Configuring Folder and Filename Tags

If you are going to analyse a set of multiple text files using Leximancer,
you should consider making use of the Apply Folder Tags function.
Essentially, if the name of each file is a category of interest, or you can
group the files into folders describing categories of interest, this feature
is a simple of way to add a lot of power to your analysis.
For example:

break up a book into various chapters (one file per chapter) to allow
you to explore what topics or characters appear in each chapter,

naming letters or reports - make each file name indicate the person or
organisation who wrote the document to let you see each of these
bodies on the map,

group newspaper articles into folders by newspaper.

Leximancer can create a category called a Tag for each folder and / or file
name in your data documents. You can create multiple levels of folders
under your parent (top-level) data folder. For example, you can create a
folder for each newspaper, and under each of those a folder for each

Manual Version 4

150

month, and under each of those a folder for each journalist. You would
place the text file for each article in the appropriate folder at the bottom
of this tree. Leximancer can then create a Tag for each folder at each
level of the tree.
To enable this function, expand the Generate Concepts Settings and click
on Edit Text Preprocessing Settings to bring up the interface below. In the
Tagging Options, select Apply folder tags or Apply file tags:

In this example, the chapters comprising the Alice in Wonderland story

were loaded into Leximancer as separate documents. When the Apply File
Tags option is enabled therefore, a tag is automatically created to
represent each the chapter in the story.
Click OK to exit the Preprocessing Settings.
Run

the

Generate

Concepts

and

Generate

Thesaurus

stages

of

processing. Expand the Run Project Settings, and Edit the Concept Coding
Settings. Use the left-hand arrow button to add the file tags to the list of
Mapping Concepts:

Manual Version 4

151

Click Ok, and then click on Run Project in main Control Panel to complete
the final phase of processing. Click on Concept Map when the project is
complete.
The concept map now includes the chapter file tags, and the concepts are
clustered around these according to their relationships. Concepts coming
from the content of a particular chapter will tend to settle near that
chapters file tag in the map space.
You can explore the topics characteristic of a chapter by clicking on a file
tag. A ranked list of related topics is revealed in the panel on the right.
These are the concepts that are coded into the chapter frequently:

Manual Version 4

152

In the example above, the Mock Turtle and Gryphon are clustered near
Chapter 10 file tag, indicating that they are fairly specific to this chapter.

Filename and folder tags are useful if you wish to explore similarities or
differences between various conditions. For example, placing speeches by
various politicians into folders according to the party to which they
belong will allow you to explore the views of the party, whereas placing
files into folders corresponding to different periods of time will allow you
to explore temporal differences.

Discrimination of Categories based on Semantics

Manual Version 4

153

The goal here is to generate concepts that distinguish each element of a

set of prior categories. For example, we might want to know what
attributes discriminate failing students from successful students. First,
ensure that the categories are defined either in the manual- or
automatic- tags tabs within the Concept Seeds Editor in the Generate
Concepts Settings. Be sure to delete any other concepts that are not
required for profiling. Endorse the Learn From Tags option in the
Thesaurus Settings in the Generate Thesaurus stage, so that the
categories can be used in the learning phase to generate related
concepts. Nominate how many concepts you wish to discover to profile
your categories, and use the Concepts in EACH setting to discover
concepts that distinguish your categories from one another. To create the
map, make sure the tag categories and the discovered concepts are in the
Mapping Concepts list in the Concept Coding Settings in the Run Project
stage. Also make sure that you use the Topical network mapping setting.

The final map will show the tag categories in proximity to their
discriminating discovered concepts. Be aware that the discovered
concepts in this case do not characterise the whole text. In fact they are
quite specific to the tag categories, and do not necessarily cover the
major themes of the whole data set at all.
In order to profile tags:

Select your data files as usual. In this example, the data consists of 4
folders containing speeches by members of Australian political parties
concerning the stem cell debate.

Expand the Generate Concepts Settings, and Edit the Text Processing
Settings. Use the Apply Folder Tags option to create a tag for the
speeches of each political party.

Manual Version 4

154

Edit on the Concept Seeds Settings and disable the radio button near
the top. Here we are interested in profiling the tags, and don't want
any automatic- or user-defined concepts present.

Run the Generate Concepts stage.

Expand the Generate Thesaurus Settings, and Edit the Concept Seeds
Settings open. In the Auto Tags tab you will see that a tag has been
created

for

each

data

sub-folder.

Delete

the

Nationals

and

Independents tags, so that you are only left with tags for the Labour
and Liberal parties. Click Ok.

Edit the Thesaurus Settings, and tick the Learn From Tags option. In
the Concept Profiling section, enter the number of concepts that you
would like to see on the map related to the two tags (60 in this
example). Since we want to discover concepts that distinguish the
party tags, use the Concepts in EACH operator this time:

Manual Version 4

155

Click Ok, then expand the Run Project Settings, and Edit the Concept
Coding Settings. Move the Folder tags of interest into the Mapping
Concepts list on the left (to allow them to be shown on the map).

Click Ok, then click the Run Project button in the main Project Control
Panel to complete the project, and then click Concept Map to see the
party profile map.

The concept map will contain concepts that distinguish the tags from one
another. The map below shows the clusters of concepts that contrast the
arguments made by the Australian Labour and Liberal parties on the
subject of the stem cell debate:

Manual Version 4

156

4. Extracting a Social Network

The goal here is to extract a network of actors, or names, based on their
interactions,

and

to

characterize

the

conceptual

nature

of

their

interactions. An example of this could be finding a network of

companies, employees, and cities from a set of industry reports.

Firstly, Edit the Concept Seeds Settings in Generate Concepts to extract

only names, as many names as you think are warranted. Force all the
automatic concepts to be names by setting the Percentage of Name-Like
Concepts to 100%.

Manual Version 4

157

After running Generate Concepts, tidy-up the automatic concepts in the

Edit Concept Seeds Settings in Generate Thesaurus. For instance, you
might remove any unwanted proper nouns, and add any missing names
that you want to watch.

Next, go into the Thesaurus Settings and enable the Concept Profiling
function. Select the Concepts in ANY operator, and choose to discover
one concept per prior name. You can increase the number of discovered
concepts if you want a richer map.

Run the Generate Thesaurus phase, then expand the Run Project Settings
and go into the Concept Coding Settings. Make sure the tags, names and
concepts of interest are included in the Mapping Concepts list. If not, use
the left-hand arrow to add them to the list.

The resulting map will show a network of names intertwined with

concepts that describe and mediate the relationships.

For a more traditional and constrained social network which uses

structural variables:

Edit the Concept Seeds Settings to select only names as described

above, then run this phase.

Manual Version 4

158

In the Edit Concept Seeds, manually seed some concepts as

structural variables.

Run the Thesaurus Learning phase.

Open the Concept Coding Settings, and put only the name-like
concepts in the Mapping Concepts list. Place the desired structural
concepts in the list of Required Concepts, then run the remaining
stages.

The result will be that only text segments that contain one of the
Required Concepts will be mapped. Consequently, the map will show a
network of names based on relationships that involve at least one of the
required concepts

Manual Version 4

159

5. Analysing Transcripts
Transcripts of meetings, interviews and focus groups can be analysed in
Leximancer as normal text, and if you group the interviews into files and
folders, you can use Folder Tagging (Chapter 14) to enhance your
analysis. Moreover, if your transcripts are in plain text or Microsoft Word
and are suitably formatted, Leximancer allows you to select, ignore, or
compare all the utterances of each distinct speaker. To allow the program
to identify the speaker of any text segment, they must be identified in a
certain way, and a new speaker label must be inserted whenever any new
speaker begins. The format requires dialogue markers which are at the
start of a paragraph; use upper case first letters for each constituent
term; are made up of a maximum of three terms; and end in a colon
followed by a space. For example:

Interviewer: So what does your son like to do for leisure, Susan?

Susan: Every Friday he plays uh ten pin bowling with the oldies. Hes not
bad either.
Alan: Oh yes, he excels at ten pin bowling, hes one of the better players
there.
Interviewer: And do you have any plans for travel coming up?
Susan: Yes actually were going up to Thailand on the 13th of October,
(B)s coming with us for 14 days. My daughter in law is there, and theyve
got a little boy.
Alan: Yeah, so well show you a photo of that, shes very cute. Four boys,
four grandsons, and one granddaughter.

Manual Version 4

160

Given text data in this format, Leximancer can extract the dialogue
markers as tags and identify the speaker of every subsequent sentence
until the next dialogue marker.

Manual Version 4

161

Configuring Transcript Analysis

Select your text data as usual. To create dialogue tags, expand the
Generate Concepts Settings and Edit the Text Processing Settings. Tick
Apply Dialogue Tags:

This transforms each dialogue marker in the text into a tag, which is then
inserted into each relevant sentence and displayed for you under
Autotags tab in the Edit Concept Seeds interface.
There is another setting in the Pre-process stage called the Prose Test
Threshold. If your interview text is quite colloquial and does not conform
to standard stop-word usage, set the Prose Test Threshold to 0. This
filter is most useful for prose interspersed with non-textual material,
such as web pages.

Run the Generate Concepts stage.

Manual Version 4

162

When this stage is complete, expand the Generate Thesaurus Settings

and Edit the Concept Seeds node to inspect the extracted textual concept
seeds in the Auto Concepts tab:

Then click on the Auto Tags tab to see the list of speakers identified by
Leximancer:

Manual Version 4

163

Now run the Generate Thesaurus phase.

When this is done, you can choose whose utterances you wish to analyse,
and whose you wish to leave out. You can also choose which items you
wish to see on the map. These settings can be changed by expanding the
Run Project Settings, and Editing the clicking Concept Coding Settings:

Manual Version 4

164

The Mapping Concepts list lets you chose which tags and concepts you
wish to appear on the concept map. By default, there are two items in the
list, All Concepts and All Discovered Names.

The All Concepts wildcard represents all of the concepts identified

for this project, be they automatically-discovered or user-defined.

The All Discovered Names wildcard represents only those namelike concepts discovered automatically by the software.

Full lists of the possible name-like and word-like concepts appear in the
right-hand panels so that you can choose which entries you would like to
see on the map.

If you wish to inspect the relative ownership of the textual concepts

between your speakers, select the desired speaker tags and use the left
arrow to add them to the Mapping Concepts list:

Manual Version 4

165

Since the full list of word-based concepts is of interest in this case, we

can leave the All Concepts wildcard in the Mapping Concepts list.

The Required and Kill Concepts tabs allow you to select whose utterances
you wish to analyse, and whose you wish to leave out from the analysis.
For example, if you wanted to suppress all the utterances of the
Interviewer, you would move the Interviewer speaker tag into the Kill
Concepts list. This causes all the concepts coded into questions asked by
the Interviewer to be removed from the analysis:

With these changes made, you can click Run Project to complete the final
phase of processing and produce a concept map:

Manual Version 4

166

6. Analysing Spreadsheet Data

Leximancer can effectively analyse spreadsheets containing text fields
and category fields (data in tabular format). You can analyse multiple text
fields in each record, and also include the categorical fields as variables
in your analysis. This enables very powerful text mining.

Practical: Analysing Spreadsheet Data

Your data may be in a spreadsheet application such as Microsoft Excel or
in a database application such as Access. The best spreadsheet
arrangement or layout for analysis is to have one separate response item
per column (the first line should contain the column headers), and one
respondent per row.

Save the spreadsheet in delimited text format (for example, in tab

delimited (.tsv) or comma separated variable (.csv) format).

Manual Version 4

167

Create a Leximancer project as usual, and click on Load Data. Browse and
select the spreadsheet file (the .tsv or .csv file). Drag and drop it into the
Document Set area, then click Ok.

Run the first phase of processing (Generate Concepts), then expand the
Generate Thesaurus Settings and Edit the Concept Seed Settings. You will
see the concept seeds automatically extracted by the software in the
Auto-Concepts tab:

Manual Version 4

168

You can seed your own concepts in the User Defined Concepts tab.
If you want more (or fewer) automatic concepts, just go back one node to
the Concept Seeds Settings and change the Total Number of Concepts to
be suggested by Leximancer.

If you process a spreadsheet that uses the layout described above,

Leximancer will automatically create a tag to represent each free-text
column, and a tag for each of the levels (or possible responses) to the
categorical variables in the data (to a limit of 500 unique responses).
Entries that read @none represent null entries or empty response cells.

Manual Version 4

169

You can review what tags Leximancer has extracted by clicking on the
Auto Tags tab. The tags take their names from the column headers and
categorical responses in the data. There is no hard limit to the number of
free-text and categorical variables that can be analysed in Leximancer.
The Auto Tags can be used for data mining correlations with textual
concepts, and for selecting which text column(s) you wish to map at any
time:

Run the Generate Thesaurus phase to extract a thesaurus from the data
describing each concept seed.

Manual Version 4

170

Expand the Run Project Settings, and Edit the Concept Coding Settings to
access the data mining options.

The Mapping Concepts list lets you select what variables you want on the
concept map, like choosing the columns you want in a database query.
For example, lets say you wish to examine the correlations of the three
position types (faculty members, staff and students) with the comments
from the text column called IT feedback comments in the data. You
would like all the textual concepts on the map, so retain the All Concepts
wildcard in the Mapping Concepts list. You would also need to select and
add the three respondent type tags (Position: Faculty, Position: Staff and
Position: Student) to the Mapping Concepts tab using the left arrow:

Once you have configured the data mining settings, run the last phase of
processing and inspect the resulting concept map:
Manual Version 4

171

You could choose to correlate the textual concepts with the satisfaction
ratings instead of the position categories. To do so, return to the data
mining settings in the Concept Coding Settings, and remove the position
tags from the Mapping Concepts list. Replace them with the satisfaction
score tags. Rerun the final phase to produce this new view of the data
very quickly:

Manual Version 4

172

You could also aggregate the satisfaction tags in the Edit Concept Seeds
stage to produce Low and High satisfaction tags to place on the
concept map. For instance, you could Merge the satisfaction scores of
1-3 and Edit this to rename it as tag as Low. Then Merge the 4-7 tags
and Edit to rename them as High. Adding the Low and High tags to
the Mapping Concepts list in the Concept Coding Settings produces the
following map:

Manual Version 4

173

You can also use the Required Concepts and Kill Concepts tabs in the
Concept Coding Settings to filter records in or out of your analysis. For
instance, if you had more than one text column in your spreadsheet, you
could choose to examine the concepts associated with a particular text
response. You would do this by moving the tag denoting the text column
of interest into the Required Concepts tab. In this case, if a data cell does
not come from that text column, it will not be coded for concepts and
mapped.
A Kill Concept is almost the opposite of a Required Concept. If a text
segment matches a Kill Concept tag or concept, it will not be coded with
any other classifier. For example, you could suppress the analysis of all
text segments which match the concept available by identifying
available as a Kill Concept.
By way of example, if we wished to map only the comments made by
women in this spreadsheet, we could either add the Gender: female tag to

Manual Version 4

174

the Required Concepts tab, or by add the Gender: male tag to the Kill
Concepts tab:

If we remove all the tags from the Mapping Concepts tab (so that only the
default All Concepts and All Discovered Names wildcards remain), the
resulting map reflects all the responses made by women in the data:

Manual Version 4

175

This concludes the Leximancer 4 Manual. Visit our

website at www.leximancer.com

*********************************************************************
This documentation is Copyright 2011 Leximancer Pty Ltd,
http://www.leximancer.com/.
All rights reserved.
********************************************************************

Manual Version 4

176