Department of Computer Science and Engineering (AIML)

(R18)
Information Retrieval System
Lecture Notes

B. Tech III YEAR – I SEM

Prepared by

Mrs.Swapna
( Professor&HOD-CSM)
Dept. CSE(AIML)

Faculty Name : Mrs Swapna Subject Name :IRS

 Syllabus

Faculty Name : Mrs Swapna Subject Name :IRS

 INFORMATION RETRIEVAL SYSTEMS (Professional Elective – II)

B.Tech. III Year I Sem. L T P C

3 0 0 3
Prerequisites: Data Structures

Course Objectives:
 To learn the important concepts and algorithms in IRS
 To understand the data/file structures that are necessary to design, and implement information retrieval (IR) systems.

Course Outcomes:
 Ability to apply IR principles to locate relevant information large collections of data
 Ability to design different document clustering algorithms
 Implement retrieval systems for web search tasks.
 Design an Information Retrieval System for web search tasks.

UNIT - I
Introduction to Information Retrieval Systems: Definition of Information Retrieval System, Objectives of Information Retrieval Systems, Functional
Overview, Relationship to Database Management Systems, Digital Libraries and Data Warehouses.
Information Retrieval System Capabilities: Search Capabilities, Browse Capabilities, Miscellaneous Capabilities.

UNIT - II
Cataloging and Indexing: History and Objectives of Indexing, Indexing Process, Automatic Indexing, Information Extraction.
Data Structure: Introduction to Data Structure, Stemming Algorithms, Inverted File Structure, N-Gram Data Structures, PAT Data Structure, Signature File
Structure, Hypertext and XML Data Structures, Hidden Markov Models.

UNIT - III
Automatic Indexing: Classes of Automatic Indexing, Statistical Indexing, Natural Language, Concept Indexing, Hypertext Linkages.
Document and Term Clustering: Introduction to Clustering, Thesaurus Generation, Item Clustering, Hierarchy of Clusters.

Faculty Name : Mrs Swapna Subject Name :IRS

 UNIT - IV
User Search Techniques: Search Statements and Binding, Similarity Measures and Ranking, Relevance Feedback, Selective Dissemination of Information
Search, Weighted Searches of Boolean Systems, Searching the INTERNET and Hypertext.
Information Visualization: Introduction to Information Visualization, Cognition and Perception, Information Visualization Technologies.

UNIT - V
Text Search Algorithms: Introduction to Text Search Techniques, Software Text Search Algorithms, Hardware Text Search Systems.
Multimedia Information Retrieval: Spoken Language Audio Retrieval, Non-Speech Audio Retrieval, Graph Retrieval, Imagery Retrieval, Video Retrieval.

TEXT BOOK:
1. Information Storage and Retrieval Systems – Theory and Implementation, Second Edition, Gerald J. Kowalski, Mark T. Maybury, Springer

REFERENCE BOOKS:
1. Frakes, W.B., Ricardo Baeza-Yates: Information Retrieval Data Structures and Algorithms, Prentice Hall, 1992.
2. Information Storage & Retrieval By Robert Korfhage – John Wiley & Sons.
3. Modern Information Retrieval By Yates and Neto Pearson Education.

Faculty Name : Mrs Swapna Subject Name :IRS

 Information Storage and Retrieval

Chapter1:
Introduction to Information Retrieval Systems

Faculty Name : Mrs Swapna Subject Name :IRS

OBJECTIVES

� Definition of Information Retrieval Systems

� Objectives of Information Retrieval Systems

� Functional Overview

� Relationship to Database Management Systems

Information Retrieval System Definition
� An Information Retrieval System is a system
that is capable of storage, retrieval, and
maintenance of information.
� Information in this context can be composed
of text (including numeric and
date data), images, audio, video and other
multi-media objects.
� Techniques are beginning to emerge to
search these other media types.
Gauge of an IR System
� An Information Retrieval System consists of a
software program that facilitates a user in finding the
information file user needs.
� The gauge of success of an information system is
how well it can minimize the overhead for a user to
find the needed information.
� Overhead from a user's perspective is tile time
required to find tile information needed, excluding
the time for actually reading the
relevant data. Thus search composition, search
execution, and reading non-relevant items are all
aspects of information retrieval overhead.
What is an Item?
� The term "item" is used to represent the smallest
complete textual unit that is processed and
manipulated by the system.
� The definition of item varies by how a specific
source treats information. A complete document,
such as a book, newspaper or magazine could be an
item. At other times each chapter, or article may be
defined as an item.
� As sources vary and systems include more complex
processing, an item may address even
lower levels of abstraction such as a contiguous passage
of text or a paragraph.
 Objectives of an IR System
� The general objective of an Information
Retrieval System is to minimize the overhead of a
user locating needed information.
� Overhead can be expressed as the time a user
spends in all of the steps leading to reading an
item containing the needed
information (e.g., query generation, query
execution, scanning results of query to select items
to read, reading non-relevant items).
Measures associates with IR systems
� Thetwo major measures
commonly associated with
information systems are
precision and recall.
• When a user decides to issue a search looking for
information on a topic, the total database is logically
divided into four segments
Measures associates with IR systems Cont.

� Relevant items are those documents that

contain information that helps the searcher in
answering his question.
� Non-relevant items are those items that do
not provide any directly useful information.
� There are two possibilities with respect to
each item: it can be retrieved or not retrieved
by the user's query.
Precision
Recall
Measures associates with IR systems Cont.

Where:
� Number_Possible_Relevant are the
number of relevant items in the database.
� Number_Total Retieved is the total
number of items retrieved from the query.
� Number_Retrieved_Relevant is the number of
items retrieved that are relevant to the user's
search need.
Measures associates with IR systems Cont.

� Precision measures one aspect of

information retrieval overhead for a user
associated with a particular search.
� If a search has a 85 per cent precision, then
15 per cent of the user effort is overhead
reviewing non-relevant items.
� Recall gauges how well a system
processing a particular query is able to
retrieve the relevant items that the user is
interested in seeing.
Ideal Precision and Recall
Ideal Precision and Recall
� Figure 1.2a shows the values of precision and recall as the number
of items retrieved increases, under an optimum query where every
returned item is relevant. There are "N" relevant
items in the database.
� In Figure 1.2a the basic properties of precision (solid line) and recall
(dashed line) can be observed.
� Precision starts off at 100 per cent and maintains that value as long
as relevant items are retrieved.
� Recall starts off close to zero and increases as long as relevant
items are retrieved until all possible relevant items have been
retrieved.
� Once all "N" relevant items have been retrieved, the only items being
retrieved are non-relevant. Precision is directly affected by retrieval
of non-relevant items and drops to a number close
to zero. Recall is not effected by retrieval of non-relevant items
and thus remains at 100 percent.
Objectives of an IR System Cont.

� The first objective of an Information Retrieval

System is support of user search generation.
� Natural languages suffer from word ambiguities
such as homographs and use of acronyms that
allow the same word to have
multiple meanings (e.g., the word "field“).
� Disambiguation techniques exist but introduce
significant system overhead in processing
power and extended search times and often require
interaction with the user.
Objectives of an IR System Cont.

� Many users have trouble in generating a good

search statement. The typical user does not
have significant experience with nor even the
aptitude for Boolean logic statements.
� Quite often the user is not an expert in the area
that is being searched and lacks domain specific
vocabulary unique to that particular subject area
(Search begins with a general
concept, a limited knowledge of the vocabulary
 associated with a particular area).

Objectives of an IR System Cont.

� Even when the user is an expert in the area

being searched, the ability to select
the proper search terms is constrained by lack
of knowledge of the author's vocabulary.
� Thus, an Information Retrieval System
must provide tools to help overcome the
search specification problems discussed above.
Vocabulary Domains
Objectives of an IR System Cont.

� An objective of an information system is to

present the search results in a format that
facilitates the user in determining relevant items.
� Historically data has been presented in an
order dictated by how it was physically
stored. Typically, this is in arrival to the
system order, thereby always displaying the
results of a search sorted by time. For those
users interested in current events
 this is useful.

Objectives of an IR System Cont.

� The new Information Retrieval Systems

provide functions that provide the results of
a query in order of potential relevance to the
user.
� Even more sophisticated techniques use item
clustering and link analysis to provide
additional item selection insights.
IR Systems Functional Overview
� A total Information Storage and Retrieval
System is composed of four major functional
processes:
1. Item Normalization,
2. Selective Dissemination of Information (i.e.,
"Mail"),
3. Archival Document Database Search, and
4. An Index Database Search.
� Commercial systems have not integrated
these capabilities into a single system but
supply them as independent capabilities.
1. Item Normalization
� Normalize the incoming items to a
standard format.
� Standardizing the input takes the different
external formats of input data and performs the
translation to the formats acceptable to
the system.
� A system may have a single format for all
items or allow multiple formats.
1. Item Normalization
Cont.
� The next process is to parse the item into logical sub-
divisions that have meaning to the user. This process,
called "Zoning," is visible to the user
and used to increase the precision of a search and
optimize the display.
� An item is subdivided into zones, which may be
hierarchical (Title, Author, Abstract, Main Text,
Conclusion, and References).
� The zoning information is passed to the processing
token identification operation to store
 the information, allowing searches to be restricted to
a specific zone.

1. Item Normalization
Cont.
� Once the standardization and zoning has
been completed, information (i.e., words)
that are used in the search process need to be
identified in the item.
� The first step in identification of a processing
token consists of determining a word. Systems
determine words by
dividing input symbols into three classes: valid
word symbols, inter-word symbols, and special
processing symbols.
1. Item Normalization
Cont.
� A word is defined as a contiguous set of
word symbols bounded by inter-word
symbols.
� Examples of word symbols are alphabetic
characters and numbers.
� Examples of possible inter-word symbols
are blanks, periods and semicolons.
1. Item Normalization
Cont.
� Next, a Stop List/Algorithm is applied to the list of
potential processing tokens.
� The objective of the Stop function is to save system
resources by eliminating from the set of searchable
processing tokens those that have
little value to the system.
� Stop Lists are commonly found in most systems and
consist of words (processing tokens) whose
frequency and/or semantic use make them of no
value as a searchable token.
� (e.g., "the"), have no search value and are not a
 useful part of a user's query.
Item Normalization Cont.

� The next step in finalizing on processing

tokens is identification of any specific word
characteristics.
� The characteristic is used in systems to
assist in disambiguation of a particular
word.
� Morphological analysis of the processing
token's part of speech is included here.
1. Item Normalization
Cont.
� Once the potential processing token has been
identified and characterized, most systems apply
stemming algorithms to normalize the token to a
standard semantic representation.
� The decision to perform stemming is a trade off
between precision of a search (i.e., finding
exactly what the query specifies) versus
standardization to reduce system overhead in
expanding a search term to similar token
representations with a potential increase in recall.
� The amount of stemming that is applied can lead to
retrieval of many non-relevant items.
2. Selective Dissemination of Information
� (Mail) Process provides tile capability to dynamically
compare newly received items in the information
system against standing statements
of interest of users and deliver the item to those users
whose statement of interest matches the contents of the
item.
� The Mail process is composed of the search
process, user statements of interest (Profiles) and
user mail files.
� When the search statement is satisfied, the item is
placed in the Mail File(s) associated with the profile.
2. Selective Dissemination of Information Cont.
� As each item is received, it is processed against
every user's profile. A profile contains a typically
broad search statement along with a list of user
mail files that will receive the document if the
search statement in the profile is satisfied.
� User search profiles are different than ad hoc
queries in that they contain significantly more
search terms (10 to 100 times more terms) and
cover a wider range of interests.
� These profiles define all the areas in which a user is
interested versus an ad hoc query which is
frequently focused to answer a specific
question.
3. Document Database Search

� The Document Database Search process is

composed of the search process, user entered
queries (typically ad hoc queries) and the document
database which contains all items that have been
received, processed and stored by the system.
� Any search for information that has already been
processed into the system can be considered a
"retrospective" search for information.
� Queries differ from profiles in that they are
typically short and focused on a specific area of
interest.
4. Index Database Search
� When an item is determined to be of
interest, a user may want to save it for
future reference. This is in effect filing it.
� In an information system this is accomplished
via the index process. In this
process the user can logically store an item in
a file along with additional index terms and
descriptive text the user wants to associate
with the item.
4. Index Database Search
� The Index Database Search Process
provides the capability to create indexes and
search them.
� The user may search the index and retrieve
the index and/or the document it references.
� The system also provides the capability to
search the index and then search the items
referenced by the index records that satisfied
the index portion of the query.
This is called a combined file search.
4. Index Database Search
� There are two classes of index files: Public
and Private Index files.
� Every user can have one or more Private Index
files leading to a very large number of files. Each
Private Index file references only a
small subset of the total number of items in the
Document Database.
� Public Index files are maintained by
professional library services personnel and
typically index every item in the Document
Database.
Relationship to Database Management Systems

� 1. An Information Retrieval System is software

that has the features and functions required to
manipulate "information" items versus a DBMS
that is optimized to handle "structured" data.
Information is fuzzy text.
� 2. Structured data is well defined data (facts)
typically represented by tables. There is a
semantic description associated with each
attribute within a table that well defines that attribute.
On the other hand, if two different people generate an
abstract for the same item, they can be different.
Relationship to Database Management Systems
� 3. With structured data a user enters a specific
request and the results returned provide the user with
the desired information. The results are
frequently tabulated and presented in a report format
for ease of use. In contrast, a search of "information"
items has a high probability of not finding all the
items a user is looking for. The user has to refine his
search to locate additional items of interest. This
process is called "iterative search.“
� From a practical standpoint, the integration of
DBMS's and Information Retrieval Systems is
very important.
Information Retrieval System Capabilities

Chapter 2
Objectives

� Discussing the major functions that

are available in an Information Retrieval
System.
� Search and browse capabilities are
crucial to assist the user in locating relevant
items.
Search Capabilities
� The objective of the search capability is to allow for a
mapping between a user's specified need and the items in
the information database that will answer
that need.
� “Weighting" of search terms holds significant potential
for assisting in the location and ranking of relevant
items.
� E.g. Find articles that discuss data mining(.9) or data
warehouses(.3).
� the system would recognize in its importance ranking and
item selection process that data mining are far more
important than items discussing data warehouses.
1. Boolean Logic
� Boolean logic allows a user to logically relate
multiple concepts together to define what information is
needed. The typical Boolean operators are AND, OR, and
NOT.

� Placing portions of the search statement in parentheses

are used to overtly specify the order of
Boolean operations (i.e., nesting function). If parentheses
are not used, the system follows a default precedence
ordering of operations (e.g.,
Use of Boolean Operators
2. Proximity
� Proximity is used to restrict the distance allowed within an item
between two search terms.
� The semantic concept is that the closer two terms are found in a text
the more likely they are related in the description of a particular
concept.
� Proximity is used to increase the precision of a search.
� If the terms COMPUTER and DESIGN are found within a few
words of each other then the item is more likely to be discussing the
design of computers than if the words are paragraphs apart.
2. Proximity
� TERM1 within "m . . . . units" of TERM2
� The distance operator "m" is an integer number and units are in
Characters, Words, Sentences, or Paragraphs.

� A special case of the Proximity operator is the Adjacent

(ADJ) operator that normally has a distance operator of one
and a forward only direction.

� Another special case is where the distance is set to zero

meaning within the same semantic unit.
2. Proximity
3. Contiguous Word Phrases
� A Contiguous Word Phrase (CWP) is both a way of
specifying a query term and a special search operator. A
Contiguous Word Phrase is two or more words that are
treated as a single semantic unit.
� An example of a CWP is "United States of America." It is
four words that specify a search term representing a single
specific semantic concept (a country) that can be
used with any of the operators discussed above.
� Thus a query could specify "manufacturing" AND "United
States of America" which returns any item that contains the
word "manufacturing" and the contiguous words
"United States of America”.
� A contiguous word phrase also acts like a special search
operator that is similar to the proximity (Adjacency) operator
but allows for additional specificity.
4. Fuzzy Searches
� Fuzzy Searches provide the capability to locate spellings of words
that are similar to the entered search term. This function is primarily
used to compensate for errors in spelling of words.
� Fuzzy searching increases recall at the expense of decreasing
precision.
� A Fuzzy Search on the term "computer" would automatically
include the following words from the information database:
"computer”, "compiter," "conputer," "computter," "compute."
� An additional enhancement may lookup the proposed
alternative spelling and if it is a valid word with a different
meaning, include it in the search with a low ranking or not
include it at all (e.g., "commuter").
� In the process of expanding a query term fuzzy searching includes
other terms that have similar spellings, giving more weight (in
systems that rank output) to words in the database
that have similar word lengths and position of the characters
as the entered term.
5. Term Masking
� Term masking is the ability to expand a query term by
masking a portion of the term and accepting as valid any
processing token that maps to the unmasked portion of
the term. The value of term masking is much higher in systems
that do not perform stemming or only provide a very simple
stemming algorithm.
� There are two types of search term masking: fixed length
and variable length.
� Fixed length masking is a single position mask. It masks out
any symbol in a particular position or the lack of that position
in a word.
� Variable length "don't cares" allows masking of any
number of characters within a processing token.
5. Term Masking
(Variable Length)
5. Term Masking
6. Numeric and Date Ranges
� Term masking is useful when applied to words, but does not
work for finding ranges of numbers or numeric dates.

� To find numbers larger than "125”, using a term "125*" will

not find any number except those that begin with the
digits "125.“

� A user could enter inclusive (e.g., "125-425" or "4/2/93-

5/2/95" for numbers and dates) to infinite ranges (">125“,
"<=233“, representing "Greater Than" or "Less Than” or
“Equal") as part of a query.
7. Concept/Thesaurus Expansion
� Associated with both Boolean and Natural Language
Queries is the ability to expand the search terms via Thesaurus or
Concept Class database reference tool.

� A Thesaurus is typically a one-level or two-level

expansion of a term to other terms that are similar in meaning.

� A Concept Class is a tree structure that expands each meaning

of a word into potential concepts that are related
to the initial term.
7. Concept/Thesaurus Expansion
7. Concept/Thesaurus Expansion
8. Natural Language Queries
� Natural Language Queries allow a user to enter a
prose statement that describes the information that the
user wants to find.

� The longer the prose, the more accurate file results

returned. The most difficult logic case associated with
Natural Language Queries is the ability to specify
negation in the search statement and have the system
recognize it as negation.
8. Natural Language Queries

� An example of a Natural Language Query is:

� Find for me all the items that discuss databases and

current attempts in database applications. Include all

items that discuss Microsoft trials in the development

process. Do not include items about relational databases.

8. Natural Language Queries
� This usage pattern is important because sentence

fragments make morphological analysis of the natural language

query difficult and may limit the system's ability to perform term

disambiguation (e.g., understand which meaning of a word is

meant).

� Natural language interfaces improve the recall of systems

with a decrease in precision when negation is required.
Browse Capabilities
� Browse capabilities provide the user with the
capability to determine which items are of interest and
select those to be displayed.
� There are two ways of displaying a summary of the
items that are associated with a query: line item status
and data visualization.
� If searches resulted in high precision, then the
importance of the browse capabilities would be
lessened.
� Since searches return many items that are not relevant to
the user's information need, browse capabilities can
assist the user in focusing on items
that have the highest likelihood in meeting his need.
1. Ranking
� Hits are retrieved in either a sorted order (e.g., sort by Title) or in
time order from the newest to the oldest item.

� With the introduction of ranking based upon predicted relevance

values, the status summary displays the relevance score associated
with the item along with a brief descriptor of the item
(usually both fit on one display screen line).

� The relevance score is an estimate of the search system on how

closely the item satisfies the search statement. Typically relevance
scores are normalized to a value between 0.0 and 1.0. The highest value
of 1.0 is interpreted that the system is sure that
the item is relevant to the search statement.
1. Ranking
� Practically, systems have a default minimum value which the
user can modify that stops returning items that have a
relevance value below the specified value.
� Presenting the actual relevance number seems to be more
confusing to the user than presenting a category that the
number falls in.
� For example, some systems create relevance categories and
indicate, by displaying items in different colors, which category an
item belongs to. Other systems uses a nomenclature such as High,
Medium High, Medium, Low, and Non-relevant. The color
technique removes the need for written indication of an
1. Ranking
� Rather than limiting the number of items that can be assessed
by the number of lines on a screen, other graphical visualization
techniques showing the relevance relationships of the hit items can
be used.
� For example, a two or three dimensional graph can be
displayed where points on the graph represent items and the location
of the points represent their relative relationship between each other
and the user's query.
� This technique allows a user to see the clustering of items by topics
and browse through a cluster or move to another topical
cluster.
2. Zoning
� The user wants to see the minimum information needed to
determine if the item is relevant.
� Limited display screen sizes require selectability of what
portions of an item a user needs to see to make the relevance
determination.
� For example, display of the Title and Abstract may be
sufficient information for a user to predict the potential
relevance of an item. Limiting the display of each item to
these two zones allows multiple items to be displayed on a
single display screen.
� This makes maximum use of tile speed of the user's
cognitive process in scanning the single image and
understanding the potential relevance of the multiple
items on the screen.
3. Highlighting
� Lets the user quickly focus on the potentially relevant
parts of the text to scan for item relevance.
� Most systems allow the display of an item to begin with the
first highlight within tile item and allow subsequent jumping
to the next highlight.
� Another capability, which is gaining strong acceptance, is for
the system to determine the passage in the document most
relevant to the query and position the browse to
start at that passage.
� Using Natural Language Processing, and automatic
expansion of terms via thesauri; highlighting loses some of its
value.
� The terms being highlighted that caused a particular item to be
returned may not have direct or obvious mapping to any of the
search terms entered.
Miscellaneous Capabilities

� There are many additional functions that facilitate the

user's ability to input queries, reducing the time it takes to

generate the queries, and reducing a priori the probability

of entering a poor query.

1. Vocabulary Browse
� The capability to display in alphabetical sorted order
words from the document database.
� The user can enter a word or word fragment and the
system will begin to display file dictionary around the
entered text.
� It helps the user determine the impact of using a fixed
or variable length mask on a search term and potential
mis-spellings.
� The user can determine that entering the search term
"compul*" in effect is searching for "compulsion" or
compulsive" or "compulsory."
2. Iterative Search and Search History Log
� The process of refining the results of a previous
search to focus on relevant items is called iterative search.

� To facilitate locating previous searches as starting

points for new searches, search history logs are available.

� The search history log is the capability to display all the

previous searches that were executed during the
current session.
3. Canned Query
� The capability to name a query and store it to be
retrieved and executed during a later user session is called
canned or stored queries.
� A canned query focuses on the user's general area
of interest one time and then retrieve it to add additional
search criteria to retrieve data that is currently needed.
� Queries that start with a canned query are
significantly larger than ad hoc queries.
 UNIT-2

CATALOGING AND INDEXING

• INDEXING: the transformation from received item to searchable data structure is called
indexing.
• Process can be manual or automatic.
• Creating a direct search in document data base or indirect search through index files.
• Concept based representation: instead of transforming the input into a searchable format
some systems transform the input into different representation that is concept based .Search ?
Search and return item as per the incoming items.
• History of indexing: shows the dependency of information processing capabilities on manual
and then automatic processing systems .
• Indexing originally called cataloguing : oldest technique to identity the contents of items to
assist in retrieval.
• Items overlap between full item indexing , public and private indexing of files
• Objectives : the public file indexer needs to consider the information needs of all users of
library system . Items overlap between full item indexing , public and private indexing of
files

• Users may use public index files as part of search criteria to increase recall.

• They can constrain there search by private index files

• The primary objective of representing the concepts within an item to facilitate users finding
relevant information .
• Users may use public index files as part of search criteria to increase recall.

• They can constrain there search by private index files

• The primary objective of representing the concepts within an item to facilitate users finding
relevant information .

Faculty Name : Mrs Swapna Subject Name : IRS

 Fig : Indexing process
• 1.Decide the scope of indexing and the level of detail to be provided. Based on usage
scenario of users .
• 2.Second decision is to link index terms together in a single index for a particular concept.

TEXT PROCESSING

1. Document Parsing. Documents come in all sorts of languages, character sets, and formats;
often, the same document may contain multiple languages or formats, e.g., a French email
16

Smartzworld.com jntuworldupdates.org
 with Portuguese PDF attachments. Document parsing deals with the recognition and
“breaking down” of the document structure into individual components. In this pre
processing phase, unit documents are created; e.g., emails with attachments are split into one
document representing the email and as many documents as there are attachments.
2. Lexical Analysis. After parsing, lexical analysis tokenizes a document, seen as an input
stream, into words. Issues related to lexical analysis include the correct identification of
accents, abbreviations, dates, and cases. The difficulty of this operation depends much on the
language at hand: for example, the English language has neither diacritics nor cases, French
has diacritics but no cases, German has both diacritics and cases. The recognition of
abbreviations and, in particular, of time expressions would deserve a separate chapter due to
its complexity and the extensive literature in the field For current approaches
3. Stop-Word Removal. A subsequent step optionally applied to the results of lexical analysis
is stop-word removal, i.e., the removal of high-frequency words. For example, given the
sentence “search engines are the most visible information retrieval applications” and a classic
stop words set such as the one adopted by the Snowball stemmer,1 the effect of stop-word
removal would be: “search engine most visible information retrieval applications”.
4. Phrase Detection. This step captures text meaning beyond what is possible with pure bag-
of-word approaches, thanks to the identification of noun groups and other phrases. Phrase
detection may be approached in several ways, including rules (e.g., retaining terms that are
not separated by punctuation marks), morphological analysis , syntactic analysis, and
combinations thereof. For example, scanning our example sentence “search engines are the
most visible information retrieval applications” for noun phrases would probably result in
identifying “search engines” and “information retrieval”.
5. Stemming and Lemmatization. Following phrase extraction, stemming and lemmatization
aim at stripping down word suffixes in order to normalize the word. In particular, stemming
is a heuristic process that “chops off” the ends of words in the hope of achieving the goal
correctly most of the time; a classic rule based algorithm for this was devised by Porter
[280]. According to the Porter stemmer, our example sentence “Search engines are the most
visible information retrieval applications” would result in: “Search engin are the most visibl
inform retriev applic”.
• Lemmatization is a process that typically uses dictionaries and morphological analysis of
words in order to return the base or dictionary form of a word, thereby collapsing its
inflectional forms (see, e.g., [278]). For example, our sentence would result in “Search
engine are the most visible information retrieval application” when lemmatized according to
a WordNet-based lemmatizer
6. Weighting. The final phase of text pre processing deals with term weighting. As
previously mentioned, words in a text have different descriptive power; hence, index terms
can be weighted differently to account for their significance within a document and/or a
document collection. Such a weighting can be binary, e.g., assigning 0 for term absence and
1 for presence.

SCOPE OF INDEXING

16

Smartzworld.com jntuworldupdates.org
 • When perform the indexing manually, problems arise from two sources the author and the
indexer the author and the indexer .
• Vocabulary domain may be different the author and the indexer.
• This results in different quality levels of indexing.
• The indexer must determine when to stop the indexing process.
• Two factors to decide on level to index the concept in a item.
• The exhaustively and how specific indexing is desired.
• Exhaustively of index is the extent to which the different concepts in the item are indexed.
• For example, if two sentences of a 10-page item on microprocessors discusses on-board
caches, should this concept be indexed
• Specific relates to preciseness of index terms used in indexing.
• For example, whether the term “processor” or the term “microcomputer” or the term
“Pentium” should be used in the index of an item is based upon the specificity decision.
• Indexing an item only on the most important concept in it and using general index terms
yields low exhaustively and specificity.
• Another decision on indexing is what portion of an item to be indexed Simplest case is to
limit the indexing to title and abstract(conceptual ) zone .
• General indexing leads to loss of precision and recall.
PREORDINATION AND LINKAGES
• Another decision on linkages process whether linkages are available between index terms for
an item .
• Used to correlate attributes associated with concepts discussed in an item .‟this process is
called preordination .
• When index terms are not coordinated at index time the coordination occurs at search time.
This is called post coordination , implementing by “AND” ing index terms .
• Factors that must be determined in linkage process are the number of terms that can be
related.
• Ex., an item discusses „the drilling of oil wells in Mexico by CITGO and the introduction of
oil refineries in Peru by the U.S.‟

AUTOMATIC INDEXING
• Case: Total document indexing

• Automatic indexing requires few seconds based on the processor and complexity of
algorithms to generate indexes.‟
• Adv. is consistency in index term selection process.

16

Smartzworld.com jntuworldupdates.org
 • Index resulting form automated indexing fall into two classes , weighted and un weighted .
• Un weighted indexing system : the existence of an index term in a document and some times
its word location are kept as part of searchable data structure .
• Weighted indexing system: a attempt is made to place a value on the index term associated
with concept in the document . Based on the frequency of occurrence of the term in the item .
• Values are normalized between 0 and 1.
• The results are presented to the user in order of rank value from highest number to lowest
number .
• Indexing By term
• Terms (vocabulary) of the original item are used as basis of index process .
• There are two major techniques for creation of index statistical and natural language.
• Statistical can be based upon vector models and probabilistic models with a special case
being Bayesian model(accounting for uncertainty inherent in the model selection process) .
• Called statistical because their calculation of weights use information such as frequency of
occurrence of words .
• Natural language also use some statistical information , but perform more complex parsing to
define the final set of index concept.
• Other weighted systems discussed as vectorised information system .
• The system emphasizes weights as a foundation for information detection and stores these
weights in a vector form.
• Each vector represents a document. And each position in a vector represent a unique
word(processing token) in a data base..
• The value assigned to each position is the weight of that term in the document.
• 0 indicates that the word was not in the document .
• Search is accomplished by calculating the distance between the query vector and document
vector.
• Bayesian approach: based on evidence reasoning( drawing conclusion from evidence )
•

• Could be applied as part of index term weighing. But usually applied as part of retrieval
process by calculating the relation ship between an item and specific query.
• Graphic representation each node represents a random variable arch between the nodes
represent a probabilistic dependencies between the node and its parents .
• Two level Bayesian network
• “ c”“ represents concept in a query

16

Smartzworld.com jntuworldupdates.org
 • Another approach is natural language processing.
• DR-LINK( document retrieval through linguistics knowledge )
• Indexing by concept
• Concept indexing determines a canonical set of concept based upon a test set of terms and
uses them as base for indexing all items. Called latent semantics indexing .
• Ex: match plus system developed by HNC inc
• Uses neural NW strength of the system word relationship (synonyms) and uses the
information in generating context vectors.
• Two neural networks are used one to generated stem context vectors and another one to
perform query.
• Interpretation is same as the weights.
• Multimedia indexing:
• Indexing video or images can be accomplished at raw data level.
• Positional and temporal (time) search can be done.

INFORMATION EXTRACTION
There are two processes associated with information extraction:
• 1.determination of facts to go into structured fields in a database and
• 2.extraction of text that can be used to summarize an item.
• The process of extracting facts to go into indexes is called Automatic File Build.
• In establishing metrics to compare information extraction, precision and recall are applied
with slight modifications.

•
Recall refers to how much information was extracted from an item versus how much should
have been extracted from the item.
16

Smartzworld.com jntuworldupdates.org
 • It shows the amount of correct and relevant data extracted versus the correct and relevant
data in the item.
• Precision refers to how much information was extracted accurately versus the total
information extracted.
• Additional metrics used are over generation and fallout.
• Over generation measures the amount of irrelevant information that is extracted.
• This could be caused by templates filled on topics that are not intended to be extracted or
slots that get filled with non-relevant data.
• Fallout measures how much a system assigns incorrect slot fillers as the number of
• These measures are applicable to both human and automated extraction processes.
• Another related information technology is document summarization.
• Rather than trying to determine specific facts, the goal of document summarization is to
extract a summary of an item maintaining the most important ideas while significantly
reducing the size.
• Examples of summaries that are often part of any item are titles, table of contents, and
abstracts with the abstract being the closest.
• The abstract can be used to represent the item for search purposes or as a way for a user to
determine the utility of an item without having to read the complete item.
DATA STRUCTURES
• Introduction to Data Structures
• Stemming Algorithms
• Inverted File Structure
• N-Gram Data Structure
• PAT Data Structure
• Signature File Structure
• Hypertext and XML Data Structures

• Data structure : The knowledge of data structure gives an insight into the capabilities
available to the system .
• Each data structure has a set of associated capabilities .
1. Ability to represent the concept and their r/s.

•
Two major data structures in any IRS:
1. One structure stores and manages received items in their normalized form is called document
manger

16

Smartzworld.com jntuworldupdates.org
 2. The other data structure contains processing tokens and associated data to support
search.

Item
normalization

Document
File creation

Document
Document
manager
Search manager

Original Major data Processing

document file
structure token

Result of a search are references to the items that satisfy the search statement which are
passed to the document manager for retrieval.
Focus : on data structure that support search function
Stemming : is the transformation often applied to data before placing it in the searchable
data structure
Stemming represents concept(word) to a canonical (authorized; recognized;
accepted)morphological (the patterns of word formation in a particular language )
representation .
Risk with stemming : concept discrimination information may be lost in the process. Causing
decrease in performance.
Advantage : has a potential to increase recall.
STEMMING ALGORITHMS
• Stemming algorithm is used to improve the efficiency of IRS and improve recall.
•

Conflation(the process or result of fusing items into one entity; fusion; amalgamation)is a

16

Smartzworld.com jntuworldupdates.org
 term that is used to refer mapping multiple morphological variants to single
representation(stem).
• Stem carries the meaning of the concept associated with the word and the affixes(ending)
introduce subtle(slight) modification of the concept.
• Terms with a common stem will usually have similar meanings, for example:
• Ex : Terms with a common stem will usually have similar meanings, for example:
• CONNECT
• CONNECTED
• CONNECTING
• CONNECTION
• CONNECTIONS
• Frequently, the performance of an IR system will be improved if term groups such as this are
conflated into a single term. This may be done by removal of the various suffixes -ED, -ING,
-ION, IONS to leave the single term CONNECT
• In addition, the suffix stripping process will reduce the total number of terms in the IR
system, and hence reduce the size and complexity of the data in the system, which is always
advantageous.
 Major usage of stemming is to improve recall.
 Important for a system to categories a word prior to making the decision to stem.
 Proper names and acronyms (A word formed from the initial letters of a name say IARE …)
should not have stemming applied.
 Stemming can also cause problems for natural language processing NPL systems by causing
loss of information .
PORTER STEMMING ALGORITHM
• Based on a set condition of the stem
• A consonant in a word is a letter other than A, E, I, O or U, some important stem conditions
are
1. The measure m of a stem is a function of sequence of vowels (V) followed by a sequence of
consonant ( C ) .
C (VC)mV. m is number VC repeats
The case m = 0 covers the null word.
2. *<X> - stem ends with a letter X
3.*v* - stem contains a vowel
4. *d - stem ends in double consonant (e.g. -TT, -SS).

16

Smartzworld.com jntuworldupdates.org
 5. *o - stem ends in consonant vowel sequence where the final consonant is
not w,x,y(e.g. -WIL, -HOP).
Suffix cond.s takes the form current _suffix = = pattern
Actions are in the form old_suffix ->. New_suffix
Rules are divided into steps to define the order for applying the rule.
Examples of the rules

Step Condition Suffix Replacement Example

1a Null Sses Ss Stresses -> stress

2.

1b *v* Ing Null Making -> mak

1b1 Null At Ate Inflated-> inflate

1c *v* Y I Happy->happi

2 m>0 aliti al Formaliti-> formal

3 m>0 Icate Ic Duplicate->duplie

4 m>1 Able Null Adjustable -> adjust

5a m>1 e Null Inflate-> inflat

5b m>1 and *d Null Single letter Control -> control

16

Smartzworld.com jntuworldupdates.org
Dictionary look up stemmers
 Use of dictionary look up.


The original term or stemmed version of the term is looked up in a dictionary and
replaced by the stem that best represents it.
 This technique has been implemented in INQUERY and Retrieval ware systems-
 INQUERY system uses the technique called Kstem.
 Kstem is a morphological analyzer that conflates words variants to a root form.
 It requires a word to be in the dictionary
 Kstem uses 6 major data files to control and limit the stemming process.
1. Dictionary of words (lexicon)
2. Supplemental list of words for dictionary
3. Exceptional list of words that should retain a „e‟ at the end (e.g., “suites” to “suite” but
“suited” to “suit”).
4. Direct _conflation - word pairs that override stemming algorithm.
5. County_nationality _conflation ( British maps to Britain )
6. Proper nouns -- that should not be stemmed
 New words that are not special forms (e.g., dates, phone numbers) are located in the
dictionary to determine simpler forms by stripping off suffixes and respelling plurals as
defined in the dictionary.
3. Successor stemmers:
 Based on length of prefixes .
 The smallest unit of speech that distinguishes on word from another
 The process uses successor varieties for a word .
Uses information to divide a word into segments and selects on of the segments to stem.

Successor variety of words are used to segment a word by applying one of the following four
16

Smartzworld.com jntuworldupdates.org
 methods.
1. Cutoff method : a cut of value is selected to define the stem length.

2. Peak and plateau: a segment break is made after a character whose successor variety exceeds
that of the character.
3. Complete word method: break on boundaries of complete words.
4. Entropy method:uses the distribution method of successor variety letters.
1. Let |Dak| be the number of words beginning with k length sequence of letters a.
2. Let |Dakj| be the number of words in Dak with successor j.
3. The probability that a member of Dak has the successor j is given as |Dakj| / |Dak|
The entropy of |Dak| is
26
Hak =  -(|Dakj| / |Dak| )(log(|Dakj| / |Dak| ))
p=1
After a word has been segmented the segment to be used as stem must be selected.
Hafer and Weiss selected the following rule
If ( first segment occurs in <=12 words in database)
First segment is stem
Else (second segment is stem)
INVERTED FILE STRUCTURE
Inverted file structure
Most common data structure
Inverted file structures are composed of three files
1. The document file
2. The inversion list (Posting List)
3. Dictionary
 The inverted file : based on the methodology of storing an inversion of documents.
 For each word a listof documents in which the word is found is stored(inversion of document
)
 Each document is given a unique the numerical identifier that is stored in inversion list .
Dictionary is used to located the inversion list for a particular word.
Which is a sorted list( processing tokens) in the system and a pointer to the location of
its inversion list.
16

Smartzworld.com jntuworldupdates.org
 Dictionary can also store other information used in query optimization such as length of
inversion lists to increase the precision.

 Use zoning to improve precision and Restrict entries.

 Inversion list consists of document identifier for each document in which the word is
found.
Ex: bit 1(10),1(12) 1(18) is in 10,12, 18 position of the word bit in the document #1.
 When a search is performed, the inversion lists for the terms in the query are locate and
appropriate logic is applied between inversion lists.
 Weights can also be stored in the inversion list.
 Inversion list are used to store concept and their relationship.
 Words with special characteristics can be stored in their own dictionary. Ex: Date… which
require date ranging and numbers.
 Systems that support ranking are re-organized in ranked order.
 B trees can also be used for inversion instead of dictionary.
 The inversion lists may be at the leaf level or referenced in higher level pointers.
 A B-tree of order m is defined as:
 A root node with between 2 and 2m keys
 All other internal nodes have between m and 2m keys
 All keys are kept in order from smaller to larger.
 All leaves are at the same level or differ by at most one level.

16

Smartzworld.com jntuworldupdates.org
Smartworld.asia Specworld.in

N-GRAM DATA STRUCTURE

• N-Grams can be viewed as a special technique for conflation (stemming) and as a unique
data structure in information systems.
• N-Grams are a fixed length consecutive series of “n” characters.
• Unlike stemming that generally tries to determine the stem of a word that represents the
semantic meaning of the word, n-grams do not care about semantics.
• The searchable data structure is transformed into overlapping n-grams, which are then used
to create the searchable database.
• Examples of bigrams, trigrams and pentagrams for the word phrase “sea colony.”
se ea co ol lo on ny Bigrams (no interword symbols)
sea col olo lon onyTrigrams (no interword symbols)
#se sea ea# #co col olo lon ony ny# Trigrams
(with interword symbol #)
#sea# #colo colon olony lony# Pentagrams (with
interword symbol #)
• The symbol # is used to represent the interword symbol which is anyone of a set of symbols
(e.g., blank, period, semicolon, colon, etc.).
 The symbol # is used to represent the interword symbol which is anyone of a set of symbols
(e.g., blank, period, semicolon, colon, etc.).
 Each of the n-grams created becomes a separate processing tokens and are searchable.
 It is possible that the same n-gram can be created multiple times from a single word.
• Uses :

16

Smartzworld.com jntuworldupdates.org
  Widely used as cryptography in world war II
 Spelling errors detection and correction

Use bigrams for conflating terms.

 N-grams as a potential erroneous words.
 Damerau specified 4 categories of errors:
Error Category Example
single char insertion compuuter
single char deletion compter
single char substitution compiter
Transposition of 2 adjacent comptuer chars
 Zamora showed trigram analysis provided a viable data structure for identifying misspellings
and transposed characters.
 This impacts information systems as a possible basis for identifying potential input errors for
correction as a procedure within the normalization process.
 Frequency of occurrence of n-gram patterns can also be used for identifying the language of
an item.
 Trigrams have been used for text compression and to manipulate the length of index terms.
 To encode profiles for the Selective Dissemination of Information.
 To store the searchable document file for retrospective search databases.

Advantage:
They place a finite limit on the number of searchable token
MaxSeg n=()n maximum number of unique n grams that can be generated.
“ n” is the length of n-grams
 number of process able symbols
Disadvantage: longer the n gram the size of inversion list increase.
Performance has 85 % precision .

PAT data structure (practical algorithm to retrieve information coded in alphanumeric)

 PAT structure or PAT tree or PAT array : continuous text input data structures(string like N-
Gram data structure).
 The input stream is transformed into a searchable data structure consisting of substrings, all
16

Smartzworld.com jntuworldupdates.org
 substrings are unique.
 Each position in a input string is a anchor point for a sub string.


In creation of PAT trees each position in the input string is the anchor point for a sub-string
that starts at that point and includes all new text up to the end of the input.
 Binary tree, most common class for prefix search,But Pat trees are sorted logically which
facilitate range search, and more accurate then inversion file .
 PAT trees provide alternate structure if supporting strings search.

Text Economics for Warsaw is complex.

sistring 1 Economics for Warsaw is complex.

sistring 2 conomics for Warsaw is complex.
sistring 5 omics for Warsaw is complex.
sistring 10 for Warsaw is complex.
sistring 20 w is complex.
sistring 30 ex.
Examples of sistrings

• The key values are stored at the leaf nodes (bottom nodes) in the PAT Tree.
• For a text input of size “n” there are “n” leaf nodes and “n-1” at most higher level nodes.
• It is possible to place additional constraints on sistrings for the leaf nodes.
• If the binary representations of “h” is (100), “o” is (110), “m” is (001) and “e” is (101) then
the word “home” produces the input 100110001101.....Using the sistrings.

INPUT 100110001101
sistring 1 1001....
sistring 2 001100...
sistring 3 01100....
sistring 4 11.......
sistring 5 1000...

16

Smartzworld.com jntuworldupdates.org
 sistring 6 000.....
sistring 7 001101
sistring 8 01101

The full PAT binary tree is

The value in the intermediate nodes (indicated by rectangles) is the number of bits to skip
until the next bit to compare that causes differences between similar terms.

Skipped final version of PAT tree

16

Smartzworld.com jntuworldupdates.org
 Signature file structure
 The coding is based upon words in the code.
 The words are mapped into word signatures .
 A word signature is fixed length code with a fixed number of bits set to 1.
 The bit positions that are set to one are determined via a hash function of the word.
 The word signatures are Ored together to create signature of an item..
 Partitioning of words is done in block size ,Which is nothing but set of words, Code length is
16 bits .
 Search is accomplished by template matching on the bit position .
 Provide a practical solution applied in parallel processing , distributed environment etc.

• To avoid signatures being too dense with “1”s, a maximum number of words is
specified and an item is partitioned into blocks of that size.
• The block size is set at five words, the code length is 16 bits and the number of bits that are
allowed to be “1” for each word is five.
• TEXT: Computer Science graduate students study (assume block size is five words)

WORD Signature

computer 0001 0110 0000 0110

Science 1001 0000 1110 0000
graduate 1000 0101 0100 0010
students 0000 0111 1000 0100
study 0000 0110 0110 0100

Block Signature 1001 0111 1110 0110

Superimposed Coding

16

Smartzworld.com jntuworldupdates.org
 Application(s)/Advantage(s)
• Signature files provide a practical solution for storing and
locating information in a number of different situations.

Hidden Markov Model:

HMM is probabilistic model for machine learning. It is mostly used in speech recognition, to
some extent it is also applied for classification task

A sequence classifier or sequence labeler is a model whose job is to assign some labelor class to
each unit in a sequence. The HMM is probabilistic sequence classifiers; which means given a
sequence of units (words, letters, morphemes, sentences etc) its job is to compute a probability
distribution over possible labels and choose the best label sequence.
The Hidden Markov Model is one of the most important machine learning models in speech and
language processing. In order to define it properly, we need to first introduce the Markov chain.

Markov chains and Hidden Markov. According to Jurafsky, Martin [2005] a weighted finite-state
automaton is a Models are both extensions of the finite automatawhich is based on the input
observation

a Markov chain is a special case of a weighted automaton inwhich the input sequence uniquely
determines states the automaton will go through for that input sequence. Since they can‟t
represent inherently ambiguous problems. Markov chain is only useful for assigning probabilities
to unambiguous sequences.

A Markov chain is specified by the following components:

Q = q1q2 . . .qN a set of states
A = [aij]NxN a transition probability matrix A, each aij representing theprobability of
moving from state i to state j.
q0,qend a special start state and end state which are not associated with observations.
A Markov chain embodies an important assumption about these probabilities In a first-order
Markov chain, the probability of a particular state is dependent only on the immediate previous
state,

Advantages and Disadvantages of HMM

The underlying theoretical basis is much more sound, elegant and easy tounderstand.

27

Smartzworld.com jntuworldupdates.org
 It is easier to implement and analyze.

HMM taggers are very simple to train (just need to compile counts from thetraining corpus).

relatively well (over 90% performance on named entities).

Statisticians are comfortable with the theoretical base of HMM.

Liberty to manipulate the training and verification processes.

Mathematical / theoretical analysis of the results and processes.

Incorporates prior knowledge into the architecture with good design.

Initialize the model close to something believed to be correct.

It eliminates label bias problem.

It has also been proved effective for a number of other tasks, such as speechrecognition,
handwriting recognition and sign language recognition.

Because each HMM uses only positive data, they scale well; since new wordscan be added
without affecting learnt HMMs.

Disadvantages :
In order to define joint probability over observation and label sequence HMMneeds to
enumerate all possible observation sequences.

Main difficulty is modeling probability of assigning a tag to word can be verydifficult if “words”
are complex.

It is not practical to represent multiple overlapping features and long termdependencies.

Number of parameters to be evaluated is huge. So it needs a large data set fortraining.

It requires huge amount of training in order to obtain better results.

HMMs only use positive data to train.

In other words, HMM training involvesmaximizing the observed probabilities for examples
belonging to a class. But it does not minimize the probability of observation of instances from
other classes.

27

Smartzworld.com jntuworldupdates.org
 Unit 3

Classes of Automatic Indexing:

Automatic indexing is the process of analyzing an item to extract the information to be permanently kept in an
index. This process is associated with the generation of the searchable data structures associated with an item. The
classes of automatic indexing are

1. Statistical Indexing (Probalistic weight calculation , vector weights-simple term frequency,inverse document
frequency,signal weighting,discrimination)

2. Concept Indexing

3. Natural Language

4.Hypertext Linkages

1.1 Statistical Indexing : Statistical indexing uses frequency of occurrence of events to calculate a number that is
used to indicate the potential relevance of an item.

Statistical strategies cover the broadest range of indexing techniques and are the most prevalent in commercial
systems.

The basis for a statistical approach is use of frequency of occurrence of events. The events usually are related to
occurrences of processing tokens (words/phrases) within documents and within the database.

The words/phrases are the domain of searchable values. The statistics that are applied to the event data are
probabilistic, Bayesian, vector space. The static approach stores a single statistic, such as how often each word
occurs in an item, that is used in generating relevance scores after a standard Boolean search.

1.2 Probabilistic indexing :stores the information that are used in calculating a probability that a particular item
satisfies (i.e., is relevant to) a particular query.

There are many different areas in which the probabilistic approach may be applied.

The method of logistic regression is described as an example of how a probabilistic approach is applied to
information retrieval

The approach starts by defining a "Model 0" system which exists before specific probabilistic models are applied. In
a retrieval system there exist query terms qi and document terms di, which have a set of attributes (vl ..... Vn) from
the query (e.g., counts of term frequency in the query), from the document (e.g., counts of term frequency in the
document ) and from the database (e.g., total number of documents in the database divided by the number of
documents indexed by the term).

The logistic reference model uses a random sample of query-documentterm triples for which binary relevance
judgments have been made from a training
1.3 Vector Weighting: A vector is a one-dimensional set of values, where the order/position of each value in the
set is fixed and represents a particular domain. In information retrieval, each position in the vector typically
represents a processing token.

There are two approaches to the domain of values in the vector: binary and weighted.

Figure 5.2 shows how an item that discusses petroleum refineries in Mexico would be represented. In the example,
the major topics discussed are indicated by tile index terms for each column (i.e., Petroleum, Mexico, Oil, Taxes,
Refineries and Shipping).

Binary vectors require a decision process to determine if the degree that a particular processing token represents
the semantics of an item is sufficient to include it in the vector.

In the example for Figure 5.2, a five-page item may have had only one sentence like "Standard taxation of the
shipment of the oil to refineries is enforced." For the binary vector, the concepts of "Tax" and "Shipment" are below
the threshold of importance (e.g., assume threshold is 1.0)
1.3.1 Simple Term Frequency Algorithm:

The simplest approach is to have the weight equal to the term frequency. This approach emphasizes the use of a
particular processing token within an item
1.3.2 Inverse Document Frequency:

weighting algorithms that the weight assigned to an item should be inversely proportional to the frequency of
occurrence of an item in the database.

This algorithm is called inverse document frequency (IDF).

The un-normalized weighting formula is:

WEIGHTij = TFij * [Log2(n) - Log2(IFj ) + 1]

where WEIGHTij is the vector weight that is assigned to term "j" in item "i," TFij (term frequency) is the frequency of
term "j" in item 'T', "n" is the number of items in the database and IFj (item frequency or document frequency) is the
number of items in the database that have term "j" in them.
1.3.3 Signal Weighting

The distribution of the frequency of processing tokens within an item can affect the ability to rank items.

For example, assume the terms "SAW" and "DRILL" are found in 5 items with the following frequencies defined in

Figure 5.5. Both terms are found a total of 50 times in the five items. The term "SAW" does not give any insight into
which item is more likely to be relevant to a search of "SAW".
1.3.4 Discrimination Value:

weighting algorithm is to base it upon the discrimination value of a term.

To achieve the objective of finding relevant items, it is important that the index discriminates among items.

There are three possibilities with the DISCRIMi value being positive, close to zero or negative.

A positive value indicates that removal of term "i" has increased the similarity between items. In this case, leaving
the term in the database assists in discriminating between items and is of value.
A value close to zero implies that the term's removal or inclusion does not change the similarity between items.

If the value of DISCRIMts is negative, the term's effect on the database is to make the items appear more similar
since their average similarity decreased with its removal.

Once the value of DISCRMi is normalized as a positive number, it can be used in tile standard weighting formula as:
Weightik = TFik * DISCRIMk

Problems With Weighting Scheme:

Dynamic changing weights when ever new data terms are entered in database becomes complex

Complex calculation for every new term entered the weightes tends to change

Large number of storage space is needed for calculation and maintainance is difficult

Problems With the Vector Model:

1. Vector representation is done by o and 1 for weighted and non weighted terms and when proximity search is
applied more number of irreverent terms may retrieve through vector model

2. Complex weight calculation for frequent changing term frequency.

1.3.5 Bayesian Model:

The objective of creating the index to an item is to represent the semantic information in the item. A Bayesian
network can be used to determine the final set of processing tokens (called topics) and their weights. Figure 5.6
shows a simple view of the process where Ti represents the relevance of topic "i" in a particular item and PTj
represents a statistic associated with the event of processing token "j" being present in the item

Figure 5.7 shows the extended Bayesian network. Extending the network creates new processing tokens for those
cases where there are dependencies between processing tokens
2 Concept Indexing:

Concept indexing maps to the specified concept for given term search according to the term weights being indexed
for specified related concept.

3.Natural Language Processing:

Natural Language Processing (NLP) refers to AI method of communicating with an intelligent systems using a natural
language such as English.
Processing of Natural Language is required when you want an intelligent system like robot to perform as per your
instructions, when you want to hear decision from a dialogue based clinical expert system, etc.

The field of NLP involves making computers to perform useful tasks with the natural languages humans use. The
input and output of an NLP system can be −

Speech

Written Text

It is the process of producing meaningful phrases and sentences in the form of natural language from some internal
representation.

It involves −

Text planning − It includes retrieving the relevant content from knowledge base.

Sentence planning − It includes choosing required words, forming meaningful phrases, setting tone of the sentence.

Text Realization − It is mapping sentence plan into sentence structure.

The NLU is harder than NLG.

Difficulties in NLU

NL has an extremely rich form and structure.

It is very ambiguous. There can be different levels of ambiguity −

Lexical ambiguity − It is at very primitive level such as word-level.

For example, treating the word “board” as noun or verb?

Syntax Level ambiguity − A sentence can be parsed in different ways.

For example, “He lifted the beetle with red cap.” − Did he use cap to lift the beetle or he lifted a beetle that had red
cap?

Referential ambiguity − Referring to something using pronouns. For example, Rima went to Gauri. She said, “I am
tired.” − Exactly who is tired?

One input can mean different meanings.

Many inputs can mean the same thing.

2.1 NLP Terminology

Phonology − It is study of organizing sound systematically.

Morphology − It is a study of construction of words from primitive meaningful units.

Morpheme − It is primitive unit of meaning in a language.

Syntax − It refers to arranging words to make a sentence. It also involves determining the structural role of words in
the sentence and in phrases.
Semantics − It is concerned with the meaning of words and how to combine words into meaningful phrases and
sentences.

Pragmatics − It deals with using and understanding sentences in different situations and how the interpretation of
the sentence is affected.

Discourse − It deals with how the immediately preceding sentence can affect the interpretation of the next
sentence.

World Knowledge − It includes the general knowledge about the world.

Steps in NLP

There are general five steps −

Lexical Analysis − It involves identifying and analyzing the structure of words. Lexicon of a language means the
collection of words and phrases in a language. Lexical analysis is dividing the whole chunk of txt into paragraphs,
sentences, and words.

Syntactic Analysis (Parsing) − It involves analysis of words in the sentence for grammar and arranging words in a
manner that shows the relationship among the words. The sentence such as “The school goes to boy” is rejected by
English syntactic analyzer.

Semantic Analysis − It draws the exact meaning or the dictionary meaning from the text. The text is checked for
meaningfulness. It is done by mapping syntactic structures and objects in the task domain. The semantic analyzer
disregards sentence such as “hot ice-cream”.

Discourse Integration − The meaning of any sentence depends upon the meaning of the sentence just before it. In
addition, it also brings about the meaning of immediately succeeding sentence.

Pragmatic Analysis − During this, what was said is re-interpreted on what it actually meant. It involves deriving those
aspects of language which require real world knowledge.

2.2 Index Phrase Generation The goal of indexing is to represent the semantic concepts of an item in the
information system to support finding relevant information. Single words have conceptual context, but frequently
they are too general to help the user find the desired information. Term phrases allow additional specification mid
focusing of the concept to provide better precision and reduce the user's overhead of retrieving non-relevant items.
Having the modifier "grass" or "magnetic" associated with the term "field" clearly disambiguates between very
different concepts. One of the earliest statistical approaches to determining term phrases proposed by Salton was
use of a COHESION factor between terms

COHESIONk.h = SIZE-FACTOR * (PAIR-FREQk.b / TOTFk * TOTFH)

where SIZE-FACTOR is a normalization factor based upon the size of the vocabulary and PAIR-FREQ~h is the total
frequency of co-occurrence of the pair Termk, Termh in the item collection.

4.HyperText Linkages:

In the context of search indexing, metadata is data that tells search engines what a webpage is about. Often the
meta title and meta description are what will appear on search engine results pages, as opposed to content from
the webpage that's visible to users.

Hyperlink Vs Hypertext

Both the term are twins to each other and perform with each other basically complete each other. But few times
we get confused about which one is the which one really. To clear that confusion, we will discuss the specificness
of both in detail with the proper example and explain the differences as well. Both the terms are used in the
WWW(World Wide Web)

Hyperlink: The hyperlink contains the URL of the webpages. In a general way, a hyperlink is referenced when a
hypertext navigated. These hyperlinks are hidden under the text, image, graphics, audio, video, and gets highlighted
once we hover the mouse over it. To activate the hyperlink, we click the hypermedia, which ends up within the
opening of the new document. It establishes the connection between the knowledge units, usually known as the
target document and therefore the alternate name for the hyperlink is anchor or node.

Hypertext: Ted Nelson introduced the term Hypertext in 1956. Hypertext is a text which contains the visible text to
redirect the targeted page(page URL contained by Hyperlink). It was invented to establish cross-reference in the
computer world, similar to that is made in books like an index. However, the usual pattern of reading a book is
sequential. But, this hypertext introduces the idea of cross-referencing the data. This cross-referencing is sort of
complicated within the world, but it makes the work easier. If we are surfing on the web, at the time of reading a
piece of writing we suddenly encounter a term, which we wanted to understand at that moment. If that term may
be a hypertext, we will directly attend that page where we will find the information about that term. So, this
eliminates the additional time of searching that term.
Example: This example combines both the term.
What is a web crawler bot?

A web crawler, spider, or search engine bot downloads and indexes content from all over the Internet. The goal of
such a bot is to learn what (almost) every webpage on the web is about, so that the information can be retrieved
when it's needed. They're called "web crawlers" because crawling is the technical term for automatically accessing a
website and obtaining data via a software program.

These bots are almost always operated by search engines. By applying a search algorithm to the data collected by
web crawlers, search engines can provide relevant links in response to user search queries, generating the list of
webpages that show up after a user types a search into Google or Bing (or another search engine).

A web crawler bot is like someone who goes through all the books in a disorganized library and puts together a card
catalog so that anyone who visits the library can quickly and easily find the information they need

How do web crawlers work?

The Internet is constantly changing and expanding. Because it is not possible to know how many total webpages
there are on the Internet, web crawler bots start from a seed, or a list of known URLs. They crawl the webpages at
those URLs first. As they crawl those webpages, they will find hyperlinks to other URLs, and they add those to the list
of pages to crawl next.

Given the vast number of webpages on the Internet that could be indexed for search, this process could go on
almost indefinitely. However, a web crawler will follow certain policies that make it more selective about which
pages to crawl, in what order to crawl them, and how often they should crawl them again to check for content
updates.

The relative importance of each webpage: Most web crawlers don't crawl the entire publicly available Internet and
aren't intended to; instead they decide which pages to crawl first based on the number of other pages that link to
that page, the amount of visitors that page gets, and other factors that signify the page's likelihood of containing
important information.

The idea is that a webpage that is cited by a lot of other webpages and gets a lot of visitors is likely to contain high-
quality, authoritative information, so it's especially important that a search engine has it indexed – just as a library
might make sure to keep plenty of copies of a book that gets checked out by lots of people.

Revisiting webpages: Content on the Web is continually being updated, removed, or moved to new locations. Web
crawlers will periodically need to revisit pages to make sure the latest version of the content is indexed.

PART 2 –unit 3

1.Introduction to Clustering

The goal of the clustering was to assist in the location of information. Clustering of words originated with the
generation of thesauri. Thesaurus, coming from the Latin word meaning “treasure,” is similar to a dictionary in that it
stores words. Instead of definitions, it provides the synonyms and antonyms for the words. Its primary purpose is to
assist authors in selection of vocabulary. The goal of clustering is to provide a grouping of similar objects (e.g., terms
or items) into a “class” under a more general title. Clustering also allows linkages between clusters to be specified.
The term class is frequently used as a synonym for the term cluster.
The process of clustering follows the following steps:

Define the domain for the clustering effort. Defining the domain for the clustering identifies those objects to be used
in the clustering process. Ex: Medicine, Education, Finance etc.

Once the domain is determined, determine the attributes of the objects to be clustered. (Ex: Title, Place, job etc
zones)

Determine the strength of the relationships between the attributes whose co-occurrence in objects suggest those
objects should be in the same class.

Apply some algorithm to determine the class(s) to which each item will be assigned.

1.2 Thesaurus Generation

There are three basic methods for generation of a thesaurus; hand crafted, co- occurrence, and header-modifier
based. In header-modifier based thesauri term relationships are found based upon linguistic relationships. Words
appearing in similar grammatical contexts are assumed to be similar. The linguistic parsing of the document
discovers the following syntactical structures: Subject-Verb, Verb- Object, Adjective-Noun, and Noun-Noun. Each
noun has a set of verbs, adjectives and nouns that it co-occurs with, and a mutual information value is calculated for
each using typically a log function.

1.3 Manual Clustering

The art of manual thesaurus construction resides in the selection of the set of words to be included. . Care is taken to
not include words that are unrelated to the domain of the thesaurus. If a

concordance is used, other tools such as KWOC, KWIC or KWAC may help in determining useful words. A Key Word
Out of Context (KWOC) is another name for a concordance. Key Word In Context (KWIC) displays a possible term in
its phrase context. It is structured to identify easily the location of the term under consideration in the sentence. Key
Word And Context (KWAC) displays the keywords followed by their context.
In the Figure 6.1 the character “/” is used in KWIC to indicate the end of the phrase. The KWIC and KWAC are useful
in determining the meaning of homographs.

Once the terms are selected they are clustered based upon the word relationship guidelines and the interpretation
of the strength of the relationship. This is also part of the art of manual creation of the thesaurus, using the
judgment of the human analyst.

1.4 Automatic Term Clustering

There are many techniques for the automatic generation of term clusters to create statistical thesauri. When the
number of clusters created is very large, the initial clusters may be used as a starting point to generate more abstract
clusters creating a hierarchy. The basis for automatic generation of a thesaurus is a set of items that represents the
vocabulary to be included in the thesaurus. Selection of this set of items is the first step of determining the domain
for the thesaurus. The processing tokens (words) in the set of items are the attributes to be used to create the
clusters.

Implementation of the other steps differs based upon the algorithms being applied. The automated method of
clustering documents is based upon the polythetic clustering where each cluster is defined by a set of words and
phrases. Inclusion of an item in a cluster is based upon the similarity of the item's words and phrases to those of
other items in the cluster.

1.4.1 Complete Term Relation Method

In the complete term relation method, the similarity between every term pair is calculated as a basis for
determining the clusters. The easiest way to understand this approach is to consider the vector model. The vector
model is represented by a matrix where the rows are individual items and the

columns are the unique words (processing tokens) in the items. The values in the matrix represent how strongly that
particular word represents concepts in the item.
Figure 6.2 provides an example of a database with 5 items and 8 terms. To determine the relationship between
terms, a similarity measure is required. The measure calculates the similarity between two terms. In Chapter 7 a
number of similarity measures are presented. The similarity measure is not critical

where “k” is summed across the set of all items. In effect the formula takes the two columns of the two terms being
analyzed, multiplying and accumulating the values in each row. The results can be paced in a resultant “m” by “m”
matrix, called a Term-Term Matrix (Salton-83), where “m” is the number of columns (terms) in the original matrix.
This simple formula is reflexive so that the matrix that is generated is symmetric. Other similarity formulas could
produce a non- symmetric matrix.

Using the data in Figure 6.2, the Term-Term matrix produced is shown in Figure 6.3. There are no values on the
diagonal since that represents the auto correlation of a word to itself. The next step is to select a threshold that
determines if two terms are considered similar enough to each other to be in the same class. In this example, the
threshold value of 10 is used. Thus two terms are considered similar if the similarity value between them is 10 or
greater. This produces a new binary matrix called the Term Relationship matrix (Figure 6.4) that defines which terms
are similar.

A one in the matrix indicates that the terms specified by the column and the row are similar enough to be in the
same class. Term 7 demonstrates that a term may exist on its own with no other similar terms identified. In any of
the clustering processes described below this term will always migrate to a class by itself.
The final step in creating clusters is to determine when two objects (words) are in the same cluster.
There are many different algorithms available. The following algorithms are the most common: cliques,
single link, stars and connected components.

Applying the algorithm to cliques Figure 6.4, the following classes are created:

Class 1 (Term 1, Term 3, Term 4, Term 6)

Class 2 (Term 1, Term 5)

Class 3 (Term 2, Term 4, Term 6)

Class 4 (Term 2, Term 6, Term 8)

Class 5 (Term 7)

Notice that Term 1 and Term 6 are in more than one class. A characteristic of this approach is that terms
can be found in multiple classes. In single link clustering the strong constraint that every term in a class is
similar to every other term is relaxed.

The rule to generate single link clusters is that any term that is similar to any term in the cluster can be
added to the cluster. It is impossible for a term to be in two different clusters. This in effect partitions the
set of terms into the clusters. The algorithm is:

Select a term that is not in a class and place it in a new class

Place in that class all other terms that are related to it

For each term entered into the class, perform step 2

When no new terms can be identified in step 2, go to step 1.

Applying the algorithm for creating clusters using single link to the Term Relationship Matrix, Figure 6.4,
the following classes are created:

Class 1 (Term 1, Term 3, Term 4, Term 5, Term 6, Term 2, Term 8)

Class 2 (Term 7)

There are many other conditions that can be placed on the selection of terms to be clustered.

Applying Star technique method following cluster is created

Class 1 (Term 1, Term 3, Term 4, Term 5, Term 6)

Class 2(Term 2, Term 4,Term 6,Term 8)

Class 3 (Term 7)
Item Clustering : same methodology as of term clustering
1.4.2 Cluster Using Existing Clusters
2.Hierarchy of Cluster or HCAM Agglomerative clustering : is the most common type of hierarchical
clustering used to group objects in clusters based on their similarity. It’s also known
as AGNES (Agglomerative Nesting). The algorithm starts by treating each object as a singleton cluster.
Next, pairs of clusters are successively merged until all clusters have been merged into one big cluster
containing all objects. The result is a tree-based representation of the objects, named dendrogram.

Uses bottom up approach method

Representation of dendogram

Objectives of dendogram:

Reduce the overhead of search

Provide for a visual representation of the information space

Expand the retrieval of relevant it

 UNIT-IV

Page | 1 User Search Techniques: Search statements and binding, Similarity measures and ranking,
Relevance feedback, Selective dissemination of information search, weighted searches of
Boolean systems, Searching the Internet and hypertext. Information Visualization:
Introduction, Cognition and perception, Information visualization technologies.

Search Statements and Binding

Search statements are the statements of an information need generated by users to specify
the concepts they are trying to locate in items.

In generation of the search statement, the user may have the ability to weight (assign an
importance) to different concepts in the statement. At this point the binding is to the
vocabulary and past experiences of the user. Binding in this sense is when a more abstract
form is redefined into a more specific form. The search statement is the user’s attempt to
specify the conditions needed to subset logically the total item space to that cluster of items
that contains the information needed by the user.

The next level of binding comes when the search statement is parsed for use by a specific
search system.
The final level of binding comes as the search is applied to a specific database. This binding
is based upon the statistics of the processing tokens in the database and the semantics used
in the database. This is especially true in statistical and concept indexing systems.
Figure 7.1 illustrates the three potential different levels of binding. Parenthesis are used in
the second binding step to indicate expansion by a thesaurus.
Page | 2

Similarity Measures and Ranking

A variety of different similarity measures can be used to calculate the similarity between the
item and the search statement. A characteristic of a similarity formula is that the results of
the formula increase as the items become more similar. The value is zero if the items are
totally dissimilar. An example of a simple “sum of the products” similarity measure from
the example to determine the similarity between documents for clustering purposes is:

Croft and Harper (Croft-79). Croft expanded this original concept, taking into account the
frequency of occurrence of terms within an item producing the following similarity formula
(Croft-83):

where K is a constant, the frequency of term is determine by “i” and J two terms
The best values for K seemed to range between 0.3 and 0.5 default threshold

Another early similarity formula was used by Salton treated the index and the
search query as n dimensional vectors . To determine the “weight” an item has with respect
 to the search statement, the Cosine formula is used to calculate the distance between the
vector for the item and the vector for
the query:

Page | 3

where is the k is a constant set to 1 and I is the term in document and j is the term for user query search.

 similarity measures for selecting Hit items is according to the value of ranking and the output is
displayed accordingly.
 Ranking the output implies ordering the output from most likely items that satisfy the query to
least likely items rank .
 Retrieval Ware first uses indexes (inversion lists) to identify potential relevant items.
 It then applies 1. coarse grain and
2.fine grain ranking

coarse grain ranking is based on the presence of query terms within items. The coarse grain ranking is a
weighted formula that can be adjusted based on completeness, contextual evidence or variety, and
semantic distance.
Fine grain ranking considers the physical location of query terms and related words using factors of
proximity search in addition to the other three factors in coarse grain evaluation, if the related terms
and query terms occur in close proximity (same sentence or paragraph) the item is judged more relevant
Relevance Feedback: The first major work on relevance feedback was published in
1965 by Rocchio (republished in 1971: Rocchio-71). Rocchio was documenting
experiments on reweighting query terms and query expansion based upon a vector
representation of queries and items. The concepts are also found in the probabilistic model
presented by Robertson and Sparck Jones (Robertson-76). The relevance feedback concept
was that the new query should be based on the old query modified to increase the weight of
Page | 4

Terms in relevant items and decrease the weight of terms that are in non-relevant items.
This technique not only modified the terms in the original query but also allowed expansion
of new terms from the relevant items. This formula was used.
Page | 5

The factor is referred to as positive feedback because it is using the user judgments on relevant
items to increase the values of terms for the next iteration of searching if it is relevant to search .
The factor is referred to as negative if it is non relevant to the query search of user.
 Figure 7.6 gives an example of the impacts of positive and negative feedback. The filled circles
represent non- relevant items; the other circles represent relevant items. The oval represents the
items that are returned from the query. The solid box is logically where the query is initially. The
hollow box is the query modified by relevance feedback (positive only or negative only in the
Figure).
Page | 6

Selective Dissemination of Information Search

Objectives of SDI:

Some of the objectives of selective dissemination of information (SDI) services are as

follows:

1. To provide current information on a predefined area of interest.

2. To receive, scan and provide the literature / information to the right users at the right
time.
3. All the current information which is relevant to the interest of the user must be brought to
the notice of the user (notification).
4. All the relevant information which is elsewhere in the world (in English or other
languages) should be located through various sources.
5. To save the time of the user.
6. No irrelevant documents should be brought to the notice of the user. Only the selective
and relevant documents should be brought to the notice of the user.

Weighted Searches of Boolean Systems

The two major approaches to generating queries are Boolean and natural
language. Natural language queries are easily represented within statistical models and are
usable by the similarity measures discussed. Issues arise when Boolean queries are
associated with weighted index systems. Some of the issues are associated with how the
logic (AND, OR, NOT) operators function with weighted values and how weights are
associated with the query terms. If the operators are interpreted in their normal
interpretation, thay act too restrictive or too general (i.e., AND and OR operators
respectively). Salton, Fox and Wu showed that using the strict definition of the operators
will sub optimize the retrieval expected by the user Closely related to the strict definition
problem is the lack of ranking that is missing from a pure Boolean process. Some of the
early work addressing this problem recognized the fuzziness associated with mixing
Boolean and weighted systems. To integrate the Boolean and weighted systems model, Fox
and Sharat author proposed a fuzzy set approach (Fox-86). Fuzzy sets introduce the concept
of degree of membership to a set. The degree of membership for AND and OR operations
 are defined as:

Page | 7

The MMM technique was expanded by Paice (Paice-84) considering all item weights versus the
maximum/minimum approach. The similarity measure is calculated as:

Searching the INTERNET and Hypertext

The Internet has multiple different mechanisms that are the basis for
search of items. The primary techniques are associated with servers on the Internet that
create indexes of items on the Internet and allow search of them. Some of the most
commonly used nodes are YAHOO, AltaVista and Lycos. In all of these systems there are
active processes that visit a large number of Internet sites and retrieve textual data which
they index. The primary design decisions are on the level to which they retrieve data and
their general philosophy on user access. LYCOS (http://www.lycos.com) and AltaVista
automatically go out to other Internet sites and return the text at the sites for automatic
indexing (http://www.altavista.digital.com). Lycos returns home pages from each site for
automatic indexing while Altavista indexes all of the text at a site. The retrieved text is then
used to create an index to the source items storing the Universal Resource Locator (URL) to
provide to the user to retrieve an item. All of the systems use some form of ranking
 algorithm to assist in display of the retrieved items. The algorithm is kept relatively simple
using statistical information on the occurrence of words within the retrieved text
Closely associated with the creation of the indexes is the technique for
Page | 8 accessing nodes on the Internet to locate text to be indexed. This search process is also
directly available to users via Intelligent Agents.

Intelligent Agents provide the capability for a user to specify an

information need which will be used by the Intelligent Agent as it independently moves
between Internet sites locating information of interest. There are six key characteristics of
intelligent agents (Heilmann-96):
1. Autonomy - the search agent must be able to operate without interaction with a human
agent. It must have control over its own internal states and make independent decisions.
This implies a search capability to traverse information sites based upon pre-established
criteria collecting potentially relevant information.
2. Communications Ability - the agent must be able to communicate with the information
sites as it traverses them. This implies a universally accepted language defining the external
interfaces
3. Capacity for Cooperation - this concept suggests that intelligent agents need to
cooperate to perform mutually beneficial tasks.
4. Capacity for Reasoning - There are three types of reasoning scenarios (Roseler-94):
Rule-based - where user has defined a set of conditions and actions to be taken Knowledge-
based - where the intelligent agents have stored previous conditions and actions taken which
are used to deduce future actions Artificial evolution based - where intelligent agents
spawn new agents with higher logic capability to perform its objectives.
5. Adaptive Behavior - closely related to 1 and 4 characteristics and adaptive behavior
permits the intelligent agent to assess its current state and make decisions on the actions it
should take
 6. Trustworthiness - the user must trust that the intelligent agent will act on the user’s
behalf to locate information that the user has access to and is relevant to the user.

Page | 9 Information Visualization

Functions that are available with electronic display and visualization of data that were
not previously provided are:
 modify representations of data and information or the display condition
(e.g., changing color scales)
 use the same representation while showing changes in data (e.g., moving
between clusters of items showing new linkages)
 animate the display to show changes in space and time
 Create hyperlinks under user control to establish relationships between data

Information Visualization addresses how the results of a search may be optimally

displayed to the users to facilitate their understanding of what the search has provided
and their selection of most likely items of interest to read. Cognitive(the mental action or
process of acquiring knowledge and understanding through thought, experience, and the
senses) engineering derives design principles for visualization techniques from what we
know about the neural processes involved with attention, memory, imagery and
information processing of the human visual system.

Cognitive engineering results can be applied to methods of reviewing the

concepts contained in items selected by search of an information system. Visualization
can be divided into two broad classes: link visualization and attribute (concept)
visualization. Link visualization displays relationships among items. Attribute
visualization reveals content relationships across large numbers of items.
There are many areas that information visualization and presentation can help the user:
a. reduce the amount of time to understand the results of a search and
likely clusters of relevant information
 b. yield information that comes from the relationships between items versus
treating each item as independent

c. perform simple actions that produce sophisticated information search functions

Page | 10 Visualization is the transformation of information into a visual form which enables the
user to observe and understand the information.

Cognition (the mental action or process of acquiring knowledge and understanding

through thought, experience, and the senses)
Perception (the ability to see, hear, or become aware of something through the senses)

The Visualization methods of representing

Figure to respective concept are

Proximity - nearby figures are grouped

together Similarity - similar figures are
grouped together
Continuity - figures are interpreted as smooth continuous patterns rather than
discontinuous concatenations of shapes (e.g., a circle with its diameter drawn is
perceived as two continuous shapes, a circle and a line, versus two half circles
concatenated together)
Closure - gaps within a figure are filled in to create a whole (e.g., using dashed lines to
represent a square does not prevent understanding it as a square)
Connectedness - uniform and linked spots, lines or areas are perceived as a single unit
 Data Visualization Techniques: There are 10 data visualization techniques

Page | 11

1. Know Your Audience

Some stakeholders within your organization or clients and partners will be happy with a simple
pie chart, but others will be looking to you to delve deeper into the insights you’ve gathered. For
maximum impact and success, you should always conduct research about those you’re
presenting to prior to a meeting, and collating your report to ensure your visuals and level of
detail meet their needs exactly.

2. Set Your Goals

To structure your visualization efforts, create a logical narrative and drill down into the insights
that matter the most. It’s important to set a clear-cut set of aims, objectives, and goals prior to
building your management reports , graphs, charts, and additional visuals.
 By establishing your aims for a specific campaign or pursuit, you should sit down in a
collaborative environment with others invested in the project and establish your ultimate
aims .
Page | 12
3. Choose The Right Chart Type
One of the most effective data visualization methods on our list; to succeed in presenting your
data effectively, you must select the right charts for your specific project, audience, and purpose.

For instance, if you are demonstrating a change over a set of time periods with more than a small
handful of insights, a line graph is an effective means of visualization. Moreover, lines make it
simple to plot multiple series together.

For you to achieve them.

4. Take Advantage Of Color Theory

The principles of color theory will have a notable impact on the overall success of your
visualization model. That said, you should always try to keep your color scheme consistent
throughout your data visualizations, using clear contrasts to distinguish between elements (e.g.
positive trends in green and negative trends in red).
 5. Handle Your Big Data
To help you handle your big data and break it down for the most focused, logical, and digestible
visualizations possible, here are some essential tips:
Page | 13
 Discover which data is available to you and your organization, decide which is the most
valuable, and label each branch of information clearly to make it easy to separate, analyze, and
decipher.

 Ensure that all of your colleagues, staff, and team members understand where your data comes
from and how to access it to ensure the smooth handling of insights across departments.

 Keep your data protected and your data handling systems simple, digestible, and updated to
make the visualization process as straightforward and intuitive as humanly possible.

 Ensure that you use business dashboards that present your most valuable insights in one easy-to-
access, interactive space - accelerating the visualization process while also squeezing the
maximum value from your information.

6. Use Ordering, Layout, And Hierarchy To Prioritize

Following on our previous point, once you’ve categorized your data and broken it down to the
branches of information that you deem to be most valuable to your organization, you should dig
deeper, creating a clearly labelled hierarchy of your data, prioritizing it by using a system that
suits you (color-coded, numeric, etc.) while assigning each data set a visualization model or chart
type that will showcase it to the best of its ability.

Of course, your hierarchy, ordering, and layout will be in a state of constant evolution but by
putting a system in place, you will make your visualization efforts speedier, simpler, and more
successful.

7. Utilize Word Clouds And Network Diagrams

A network diagram is often utilized to draw a graphical chart of a network. This style of layout is
useful for network engineers, designers, and data analysts while compiling comprehensive
network documentation.

8. Include Comparisons
This may be the briefest of our data visualization methods, but it’s important nonetheless: when
you’re presenting your information and insights, you should include as many tangible
 comparisons as possible. By presenting two graphs, charts, diagrams together, each showing
contrasting versions of the same information over a particular timeframe, such as monthly sales
records for 2016 and 2017 presented next to one another, you will provide a clear-cut guide on
the impact of your data, highlighting strengths, weaknesses, trends, peaks, and troughs that
everyone can ponder and act upon.
Page | 14

9. Apply Visualization Tools For The Digital Age

We live in a fast-paced, hyper-connected digital age that is far removed from the pen and paper
or even copy and paste mentality of the yesteryears - and as such, to make a roaring visualization
success, you should use the digital tools that will help you make the best possible decisions while
gathering your data in the most efficient, effective way.

A task-specific, interactive online dashboard or tool offers a digestible, intuitive, comprehensive,

and interactive mean of collecting, collating, arranging, and presenting data with ease - ensuring
that your techniques have the most possible impact while taking up a minimal amount of your
time.

10. Tell Your Tale

Similar to content marketing, when you're presenting your data in a visual format with the aim of
communicating an important message or goal, telling your story will engage your audience and
make it easy for people to understand with minimal effort.

Developing the information visualization display for retrieval of information is to first

1. Identifying the subset of items that is relevant to the search statement

2. Applying a threshold to determine the subset to process for visualization

3. Calculating the pair wise similarity between all of the indicated items and clustering the
results

4. Determining the theme or subject of the clusters

 5. Determining the strength of the relationships between the clusters

6. Creating the information visualization for the results

Page | 15
 Unit 5
Text Search Algorithms: Introduction, Software text search algorithms, Hardware text
search systems.
Multimedia Information Retrieval: Spoken Language Audio Retrieval, Non-Speech Audio
Retrieval,
Graph Retrieval, Imagery Retrieval, Video Retrieval

The basic concept of a text scanning system is the ability for one or moreusers to enter queries, and
the text to be searched is accessed and compared to thequery terms. When all of the text has been
accessed, the query is complete.

The database contains the full text of the items .The term detector is the special hardware/software
that contains all of the terms being searched for and in some systems the logic between the items. It
will input the text and detect the existence of the search terms. It will output to the query resolver
the detected terms to allow for final logical processing of a query against an item. The query
resolver performs two functions.

It will accept search statements from the users, extract the logic and search terms andpass
the search terms to the detector. It also accepts results from the detector and determines which
queries are satisfied by the item and possibility the weight associated with hit. The Query Resolver
will Pass information to the user interface that will be continually updating search status to the user
and on request retrieve any items that satisfy the user search statement.
In the case of hardware search machines, multiple parallel search machines (term detectors)
may work against the same data stream allowing for more queries or against different data streams
reducing the time to access thecomplete database. In software systems, multiple detectors may
execute at the same time.
Text search Techniques are of two types
1.Hardware text search
 2. software text search

Hardware Text Search Systems

Software text search is applicable to many circumstances but has encountered restrictions on the
ability to handle many search terms simultaneously against the same text and limits due to I/O
speeds. One approach that off loaded the resource intensive searching from the main processors
was to have a specialized hardware machine to perform the searches and pass the results to the
main computer which supported the user interface and retrieval of hits. Since the searcher is
hardware based, scalability is achieved by increasing the number of hardware search devices.
Another major advantage of using a hardware text search unit is in the elimination of the
index that represents the document database. Typically the indexes are 70% the size of the actual
items. Other advantages are that new items can be searched as soon as received by the system rather
than waiting for the index to be created and the search speed is deterministic.
Figure 9.1 represents hardware as well as software text search
solutions. The arithmetic part of the system is focused on the term detector. There has been three
approaches to implementing term detectors: parallel comparators or associative memory, a cellular
structure, and a universal finite state automata.
When the term comparator is implemented with parallel comparators, each term in the
query is assigned to an individual comparison element and input

data are serially streamed into the detector. When a match occurs, the term comparator informs the
external query resolver (usually in the main computer) by setting status flags.

Specialized hardware that interfaces with computers and is used to search secondary storage
devices was developed from the early 1970s with the most recent product being the Parallel
Searcher (previously the Fast Data Finder). The typical hardware configuration is shown in
Figure 9.9 in the dashed box. The speed of search is then based on the speed of the I/O.
 One of the earliest hardware text string search units was the Rapid Search Machine
developed by General Electric. The machine consisted of a special purpose search unit where a
single query was passed against a magnetic tape containing the documents. A more sophisticated
search unit was developed by Operating Systems Inc. called the Associative File Processor (AFP).
It is capable of searching against multiple queries at the same time. Following that initial
development, OSI, using a different approach, developed the High SpeedText Search (HSTS)
machine. One state machine is dedicated to contiguous word phrases, another for imbedded term
match and the final for exact word match.

In parallel with that development effort, GE redesigned their Rapid Search Machine into the
GESCAN unit. The GESCAN system uses a text array processor (TAP) that simultaneously
matches many terms and conditions against a given text stream the TAP receives the query
information from the user’s computer and directly access the textual data from secondary storage.
The TAP consists of a large cache memory and an array of four to 128 query processors. The text is
loaded into the cache and searched by the query processors (Figure 9.10). Each query processor is
independent and can be loaded at any time. A complete query is handled by each query processor.

A query processor works two operations in parallel; matching query terms to input text and
Boolean logic resolution. Term matching is performed by a series of character cells each containing
one character of the query. A string of character cells is implemented on the same LSI chip and the
chips can be connected in series for longer strings. When a word or phrase of the query is
matched, a signal is sent to the resolution sub-process on the LSI chip. The resolution
chip is responsible for resolving the Boolean logic between terms and proximity
requirements. If the item satisfies the query, the information is transmitted to the users
computer.

The text array processor uses these chips in a matrix arrangement as shown in
Figure9.10. Each row of the matrix is a query processor in which the first chip performs
the query resolution while the remaining chips match query terms. The maximum
number of characters in a query is restricted by the length of a row while the number of
rows limit the number of simultaneous queries that can be processed.

Another approach for hardware searchers is to augment disc storage called as

Fast data finder. The augmentation is a generalized associative search element placed
between the read and write heads on the disk. The content addressable segment
sequential memory (CASSM) system uses these search elements in parallel to obtain
structured data from a database. The CASSM system was developed at the University of
Florida as a general purpose search device. It can be used to perform string searching
across the database. Another special search machine is the relational associative
processor (RAP) developed at the University of Toronto. Like CASSM performs search
across a secondary storage device using a series of cells comparing data in parallel.

The Fast Data Finder (FDF) is the most recent specialized hardware text search
unit still in use in many organizations. It was developed to search text and has been used
to search English and foreign languages. The early Fast Data Finders consisted of an
array of programmable text processing cells connected in series forming a pipeline
hardware search processor. The cells are implemented using a VSLI chip. In the TREC
tests each chip contained 24 processor cells with a typical system containing 3600 cells.
Each cell will be a comparator for a single character limiting the total number of

4
characters in a query to the number of cells.

The cells are interconnected with an 8-bit data path and approximately 20- bit
control path. The text to be searched passes through each cell in a pipeline fashion until
the complete database has been searched. As data is analyzed at each cell, the 20 control
lines states are modified depending upon their current state and the results from the
comparator. An example of a Fast Data Finder system is shown in Figure 9.11.

 A cell is composed of both a register cell (Rs) and a comparator (Cs).

 The input from the Document database is controlled and buffered by the

5
 microprocessor/memory and feed through the comparators.
 The search characters are stored in the registers.
 The connection between the registers reflects the control lines that are also passing state
information.

 Groups of cells are used to detect query terms, along with logic between the terms, by
appropriate programming of the control lines.
 When a pattern match is detected, a hit is passed to the internal microprocessor that
passes it back to the host processor, allowing immediate access by the user

The functions supported by the Fast data Finder for a query search supports:
 Boolean Logic including negation
 Proximity on an arbitrary pattern
 Term masking
 Fuzzy matching
 Term weights
 Numeric ranges

Software Text Search Algorithms

 In software streaming techniques, the item to be searched is read into memory, and then
the algorithm is applied.
 There are four major algorithms associated with software text search:
 the brute force approach,
 Knuth-Morris-Pratt(in Syllabus)
 Boyer-Moore(in Syllabus)
 Shift-OR algorithm, and Rabin-Karp

6
7
 Knuth Morris Pratt (KMP) is an algorithm, which checks the characters from left to
right.
 When a pattern has a sub-pattern appears more than one in the sub-pattern, it uses that
property to improve the time complexity, it is also applyed for the worst case.
 Input and Output
 Input: Main String: “AAAABAAAAABBBAAAAB”,
 The pattern “AAAB”
 Output:

8
 Pattern found at location: 1
 Pattern found at location: 7
 Pattern found at location: 14

9
10
The Boyer Moore algorithm does preprocessing , It processes the pattern and creates different arrays for
each of the two heuristics.
At every step, it slides the pattern by the max of the slides suggested by each of the two heuristics.
Boyer Moore is a combination of the following two approaches.
1) Bad Character Heuristic
2) Good Suffix Heuristic

11
Rabin Krap Method

12
Multimedia Information Retrieval: Spoken Language Audio Retrieval, Non-Speech
Audio Retrieval,
Graph Retrieval, Imagery Retrieval, Video Retrieval

1. Introduction

The amount of information available on the Internet, primarily by way of the World

Wide Web, is truly staggering. According to one measurement, in February 1999 there were

about 800 million web pages publicly available on about 3 million web servers, for a total of

approximately 9 terabytes of data These enormously large numbers are a testament to the

success of the Internet in providing a way for people all around the world to share information

and communicate with each other.

With computer technology improving at a phenomenal pace, the technology limitations which

dictated the predominant use of text on the Internet in the past are lessening. In the very near

future, non-textual data will be as common a format for publicly available data as text is now.

In light of these trends, it is important to review the state of the art of the retrieval of

such non- textual, multimedia data. Text information retrieval is already well established;

most data retrieval systems, such as web search engines, are text retrieval systems. However,

multimedia information retrieval is less established. There are a number of open issues

involved in such retrieval.

Image Data Retrieval

Of image, audio, and video, image data retrieval is arguably the best developed

technology. As far back as 1986, image databases were being developed and deployed, such

as UC Berkeley’s ImageQuery system, whose “developers believe that this software […] was

13
the first deployed multi-user networked digital image database system” [2]. With over a

decade of research and development, image data retrieval has had time to grow and mature.

This has allowed the area to address some difficult issues (some of which remain open at

present): image classification, query matching, image standards, attribute classification, and

evaluation. These issues will be explained further below. As a note, though standards,

attribute classification, and evaluation are discussed in terms of image retrieval systems, they

are outstanding issues for audio and video retrieval systems as well. Classification and

querying also apply to the other forms of media, but the media’s unique properties necessitate

different classification and query matching algorithms for each.

Image Classification

Image classification is concerned with assigning some higher-level semantic meaning

to the amalgamation of pixels that make up an image document. Usually the primary

motivation behind such classification is to enable query matching, which is discussed below,

but classification is a complex issue and warrants its own section. This section describes

different ways to classify images, regardless of intent. The context for the discussion is

through pattern recognition.

Image classification is primarily a pattern recognition problem. For a human being,

pattern recognition is innate and often subconscious; optical illusions, for example, play on

this fact by often inviting the eye to see patterns that are inaccurate or incorrect. Even babies

learn at an extremely early age to identify a parent’s face. For an automated image processing

system, however, pattern recognition is a surprisingly complex problem. The same level of

detail that allows computers to perform large numerical computations with unerring accuracy

14
works against computers attempting to recognize patterns in images. Since two images of the

same object can be slightly different, such as different angles of view, different lighting,

different coloring, etc., a computer’s precision does not easily “ignore” such differences.

Humans, of course, with their (relatively) larger lack of precision, can easily see past minor

variations and classify similar objects correctly.

One approach that has been reported in the literature to address the pattern recognition

problem is the general technique of segmentation. The segmentation technique is based on the

classic computer science strategy of divide-and-conquer to reduce the problem to smaller

chunks, which are easier to solve and whose solutions can be combined to eventually solve the

larger problem. In this case, the pattern recognition problem is segmented into three levels of

matching: the pixel level, the “stuff” level, and the “thing” level. The pixel level is the

computationally simplest level; the system performs basic comparisons on corresponding

pixels in the images. It is also generally the least useful technique, as minor changes in image

appearance can render a false negative. However, using pixel level matching as a basis,

higher-order matching can be performed, using queries such as “a mostly green area with some

brown vertical strips,” which could be a forest with trees. This level of recognition is the

“stuff” level, as the system now has an awareness of some relationships between pixels to

represent some stuff. Using stuff, an even higher semantic meaning can be assigned to

relationships between stuff – “things,” such as “a mostly cylindrical area with four smaller

cylinders below it, and all cylinders an alternating mix of white and black regions” to (very

crudely) represent a zebra. The “thing” level is the level in which most humans would prefer

to operate, as the semantic units are clear, discrete, and of an appropriate scale. A human

15
would normally search for all images of zebras, not all images of cylinders with smaller

cylinders below it, where all cylinders have patterns of alternating black and white. This

segmentation into pixel, stuff, and thing levels provides a tractable approach to the problem of

pattern recognition

Segmentation is merely a technique designed to address the pattern recognition

problem. An implementation of the segmentation approach is presented in The system,

Blobworld, segments an image into contiguous regions of pixels (“blobs”) which have similar

color and texture. The authors admit their blobs are not quite at the same semantic level as

“things,” but they state that blobs are semantically higher than “stuff.” Additionally, their

system provides some key features lacking in other image retrieval systems: an interface to

allow the user to sketch blobs for a query, and feedback as to why the system matched an

image with the query. Blobworld, while perhaps not yet well enough developed for general

public usage, is a promising research prototype towards solving the pattern recognition

problem through segmentation.

Query Matching

Tied very closely to the issue of image classification is the issue of query matching.

As previously stated, the primary intent behind classifying images is to allow efficient

searching or browsing to the database of images. The range of types of queries supported by

an image retrieval system will be primarily based on how the images are classified. For

example, a system that classifies its images using segmentation and generates “stuff” would

(hopefully) allow searchers to query the database based on some criteria of stuff. Clearly, any

image retrieval system can support text keyword matching based on manually indexed

16
metadata, but such querying is generic and essentially ignores the format of the image

documents. Three querying techniques that have been developed which take into account the

unique properties of image data are color histograms, quadtrees of histograms, and basic

shape matching.

Searching by color data is essentially a pixel-level search. Since pixel comparisons are

basically numerical comparisons and do not require semantic reasoning, they are very easy for

computers to perform. An example of such a query could be “find all images with at least 50%

more red pixels than green pixels” or “find all images whose most frequently used color is

similar to this image’s most frequently used color.” Searches of this type are often answered

through a color histogram, essentially a summarization of an image by the frequency of color

occurences in that image. Often histograms are stored internally as vectors of values, which

are easy to search by the matching algorithms. For example, Columbia University’s

WebSEEK system uses color histograms to keep its query response time less than two seconds

[5].

Though the computer can therefore process pixel level searches with color histograms

very quickly and efficiently, it is clear they are likely to be of very limited use to a human

searcher.

Image Standards

Image standards refers to the standards that define the metadata which describe image

files. The most obvious metadata is the structure of the electronic image file itself. Widely

adopted, open standards, such as JPEG, GIF, and TIFF, have been developed and deployed

and allow the easy sharing of images. The ready availability of the details of the standard

17
provides a measure of confidence that these file formats will be decodable even in the future.

Repositories of file format information exist (e.g., [17]), even providing decoding information

for long-obsolete formats as Wordstar and dBASE files. Given such repositories, files from

long ago could still be decoded and used, albeit with some effort.

Despite such access to file format information, however, the problem is not yet a solved

problem. Not only is image metadata important to simply understand and decode the image

document, but a large amount of other metadata needs to accompany image files for future

reference. Such metadata could include information about how the image was generated (e.g.,

a scan of a photograph of an original painting, or a digital picture of a building digitally

retouched to remove shadows). An indication of the contents of the image would also be

desirable, allowing the comparing of two images (such as two digital photographs of the same

statue taken from different angles) for a measure of equivalency. The metadata could include

information about reproduction rights of the image, or contact information for the holder of the

copyright. Finally, the metadata might include some sort of verification signature to assure the

veracity of all the metadata information, or the authenticity of the image

Audio Data Retrieval

Audio data retrieval systems are not text-based retrieval systems, and they therefore

share the same issues as image retrieval systems. As stated above, the issues of standards,

attribute classification, and evaluation are directly applicable to audio retrieval as well. They

also pose different research problems than image retrieval systems do, for two fundamental

reasons: audio data is (obviously) aurally-based instead of visually-based, and audio data is

time-dependent. The former difference leads to some unique and creative approaches to

18
solving the querying and retrieval issue, while the latter difference is the root of the

interesting problem of presentation, which image retrieval systems do not share.

Querying and Retrieval

Just as image retrieval systems must address how to support queries for images, audio

retrieval systems must create ways to allow formation of queries for audio documents.

Naturally, (text) keyword matching is a possibility, just as it can be used in image retrieval

systems. However, the natural way for humans to query other human retrieval systems (e.g.,

music librarians, radio DJs, employees at music stores, etc.) is by humming or singing part of a

tune.

Research into how non-professional singers hum or sing familiar songs has led to the

development of a number of systems which can accept such hummed or sung input for queries

After accepting the acoustic query and transforming it to digital format, there are different

ways to perform the actual matching. Bainbridge et al.’s system describes how they use

frequency analysis to transcribe the acoustic input into musical notes and then compare edit

distances to determine matches Ghias et al.’s approach differs; they convert the input into a

pitch contour, which is a string in a three- letter alphabet. The pitch contour represents how

the pitch of the input changes between each note: whether the pitch goes up (U), goes down

(D), or stays the same (S). Given this string, familiar string- comparison algorithms can be

used to determine matches against the audio database [10].

Using just three choices to generate the pitch contour means simpler matching, but it

also means that a large amount of information is discarded which could reduce the search

space. Blackburn and DeRoure suggest various improvements to the query process, including

19
a five-letter alphabet (up a lot, up a little, same, down a little, and down a lot); generating a

secondary pitch contour, where a note is compared to the note two notes ago; and comparing

time contours, which would represent rhythm information . Ghias et al. additionally note that

some errors, such as drop-out errors (skipping notes) may be more common when people hum

or sing a song. They suggest further study to clarify the relative frequency of such errors, so

as to allow tuning of the matching algorithms to be more tolerant of the common errors

The nature of audio and music data presents many opportunities to develop creative

methods to accept and process audio queries. Using error-tolerant abstractions such as

frequency analysis or pitch contours, audio retrieval systems can transform the problem of

audio matching into well-known problems of edit distance calculation or string matching. In

this way, systems can utilize established solutions for these problems to provide efficient and

effective audio retrieval.

Presentation

Of course, once the user has input a query and the system has determined some

number of matches against the audio database, the next logical step is presenting the match

results to the user. Here the time-dependent nature of audio data reveals the problem of

presentation. For media that are not time-dependent, such as text or images, the data (or an

abbreviated form) is static and can be displayed without any trouble. For time-dependent

media such as audio, it is unclear what form should be displayed or presented to the user,

since simultaneously playing 20 clips of music (representing 20 query matches displayed at a

time) is unlikely to be useful to the searcher. Bainbridge et al. enumerate a number of such

problems in presenting retrieved audio, especially when compared to typical functionality

20
supported in presenting retrieved text. These issues include whether to transpose all matches

to the same key to make comparison easier, using a visual representation to present the audio,

allowing for the equivalent of quickly scanning through a list of matches to find an

appropriate match, supporting excerpting to show the matched query in context, and creating

summaries of audio to speed relevance judgments

A related research effort is how to browse and navigate through databases of audio.

Audio is inherently a stream of time-dependent auditory data, with no standardized structure

for interconnecting related points in time in these streams. For text, hypertext provides a

structure to indicate relationships between certain parts of the text, both within the same

document and between documents. Blackburn and DeRoure describe their attempts to provide

a similar functionality for music .They propose to use an open hypermedia model to supply

hyperlinks. This model specifies that hyperlinks are not embedded in the contents document,

but instead are stored in a separate, associated document (the “linkbase”). At any point while

browsing a music document, a user may request hyperlinks based on the current location in

the audio stream; the system will then consult the linkbase to present links to related materials.

This content-based navigation is aimed at adding structure to the otherwise unstructured

streams of audio documents.

Video Data Retrieval

Video data retrieval shares some properties with image data retrieval, due to the

commonality of their visual nature. However, video data is also time-dependent like audio

data, and, in fact, movies often have synchronized audio tracks accompanying the video data.

This shared commonality naturally lends to applying solutions from the image and audio

21
retrieval areas to research problems in the video retrieval domain. In some ways this strategy

is successful, but, as usual, video data has some unique properties which again lead to creative

solutions to the research issues of classification for querying and presentation.

Classification for Querying

Some novel approaches have been developed to classify video data for good query
matching.

Gauch et al. describe how their VISION system processes video data for classification

through segmentation This segmentation is slightly different than the segmentation in terms

of image data; specifically, segmentation here means to identify camera shot changes in the

stream of video data, and from there to group adjacent camera shots into scenes. This is

analogous to segmentation of image data into stuff and things, and unfortunately, the

difficulty of such classification is analogous as well. It has been well researched how to

identify changes of camera shots, such as by observing large changes in color histograms

between frames. However, it is more complicated to properly identify when a scene starts

and ends. The VISION system uses clues from the synchronized audio track to perform this

segmentation; for example, if the speaker changes after a shot change, it may signify a

different scene. By tuning various thresholds, the VISION system can be adjusted to correctly

segment most video data.

Another processing feature of VISION is the use of the closed-captioning signal, if it

exists, to help classify the video data. Keywords are extracted from the text of the closed-

captioning, using well- understood text manipulation techniques. This provides a reliable

source of metadata information for classification. If the closed-captioning signal is absent, the

22
VISION system falls back to extracting keywords from the audio stream. They take care to

make the distinction between full continuous speech recognition of the audio stream, which is

a difficult task, to what they call “word-spotting,” or selective keyword recognition from the

audio. Gauch et al. admit their word-spotting technique does not yield very good results yet

(about 50% recall but only 20% precision), but they intend to refine and improve the method.

Another classification strategy is the use of keyframes. Keyframes are frames whose

images represent a semantic unit of the stream, such as a scene. Many video retrieval systems

implement some algorithm to identify keyframes . Color features and motion cues can be used

to automatically detect keyframes By extracting keyframes, the retrieval system can leverage

image retrieval techniques to support queries on keyframe images. Assuming the keyframes

are indeed good representatives of their respective scenes, this classification method is also a

very useful way to provide efficient browsing of the video data.

Video data is made up of both image data and audio data, and this fact provides ways to

approach the problem of classification for queries. Using video segmentation, analyzing the

closed-captioning or audio signal, and extracting keyframes are some of the ways to implement

effective video data classifiers.

Presentation

Owing to its time-dependence, video data shares audio data’s difficulty in

presentation. The distinct properties of video data, however, allow different techniques to

address this issue. As mentioned above, keyframe extraction provides the (supposedly) most

important frames in the video document, and these frames can be used as a summarization of

the entire document. WebClip, for example, calls this model the time-based model since the

23
 timeline is kept in the correct sequential order for presentation. The VISION system uses this

technique as well for its presentation, displaying thumbnails of each keyframe and showing

the full video data if the user selects a specific thumbnail They also mention that during such

playback, the user interface includes fast-forward and fast-rewind buttons, which display the

video stream at four times the normal rate (usually by dropping frames to achieve the desired

rate), and a slider bar to allow access to any arbitrary moment in the video.

Graph Based Retrieval

graph-based information retrieval system whose query can be expressed as a graph of topics/subtopics. Documents are
ranked with respect to a query, upon relationships among documents, relationships among topics/subtopics, and
relationships between query terms and documents.

The system is evaluated and compared with two information retrieval systems on two standard text collections.

24