Information Retrieval and Web Search
Salvatore Orlando
Bing Liu. Web Data Mining: Exploring Hyperlinks, Contents,
and Usage Data. Springer-Verlag, 2006
Christopher D. Manning, Prabhakar Raghavan and Hinrich
Schtze, Introduction to Information Retrieval, Cambridge
University Press. 2008
(http://nlp.stanford.edu/IR-book/information-retrieval-book.html)
Data and Web Mining. - S. Orlando 1
Introduction
Text mining refers to data mining using text
documents as data.
Most text mining tasks use Information Retrieval
(IR) methods to pre-process text documents.
These methods are quite different from traditional
data pre-processing methods used for relational
tables.
Web search also has its root in IR.
First, discuss the feature of the Web data
Data and Web Mining. - S. Orlando 2
Web
Web:
A huge, widely-distributed, highly heterogeneous, semistructured,,
interconnected, evolving, hypertext/hypermedia information
repository
Main issues
Abundance of information
The 99% of all the information are not interesting for the 99% of all
users
The static Web is a very small part of all the Web
Dynamic Website
To access the Web user need to exploit Search Engines (SE)
SE must be improved
To help people to better formulate their information needs
More personalization is needed
Data and Web Mining. - S. Orlando 3
Web: Trends and Features
Such numbers represent an estimation of minimum size of Internet.
The websites are many more, while the number of pages is almost
endless
Data and Web Mining. - S. Orlando 4
Web: Trends and Features
Google in July 2007 announced to have identified 1 trillion
(1012) of unique pages/URLs in the Web
After removing duplicates (about 30%-40%) !!!
Estimated growth: several billions of pages per day
Source: http://googleblog.blogspot.com/2008/07/we-knew-web-
was-big.html
Note that many pages are dynamically created . and this
introduces complexity for systems like Google
Think about a Web calendar on the Web . and a link to next
month we can follow this link a unbounded number of times, by
creating always new pages
Data and Web Mining. - S. Orlando 5
Web: Trends and Features
How many disks for all the
Web pages?
Consider only the text (HTML)
On the average, 10K Byte
Considering a trillion of pages
about 1016 Byte
Using Hard Disks of 1 Terabyte (about 1012 bytes)
About 10.000 disks L
Things make worse if we consider multimedia data
Data and Web Mining. - S. Orlando 6
Web: Trends and Features
Besides the grows of the page number, the pages
are also continuosly updated or removed
About the 23% of all the pages are modified daily
In the .com domain, this percentage rises to 40%
On the average, after 10 days, half of the new pages are
removed
Their URL are no longer valid
A. Arasu et al., Searching the Web, ACM Transaction on Internet Technology,
1(1), 2001. Data and Web Mining. - S. Orlando 7
Web: Trends and Features
The structure of the Web
network/graph (Bow-tie )
28% of all the pages
Core of the network
Important pages
highly interconnected
with each other
22% of all the pages
reachable from the core,
but not vice versa
The rest of the pages are disconnected from the network core
Andrei Broder, et al. Graph structure in the web: experiments and models 9th WWW, 2000.
Data and Web Mining. - S. Orlando 8
Web: Trends and Features
Power law
The degree of a node is
the number of
incoming/outgoing
links
If we call k the degree
of a node, a scale-free
network is defined by
the power-law, which
corresponds to this
distribution:
log(frequency)
80%
Most of the nodes are
poorly interconnected
log(degree)
Andrei Broder, et al. Graph structure in the web: experiments and models 9th WWW, 2000.
Data and Web Mining. - S. Orlando 9
The Power law (Long Tail) is ubiquitous in the Web
Contents
Words in the pages (frequency of words): the most common
words are very popular, but there is a long tail of infrequent
words!
Structure
In-degrees / Out-degrees / Numbers of pages per website
Usage patterns
Numbers of visitors
Query/Terms submitted by users of Search Engine
Data and Web Mining. - S. Orlando 10
Long Tail in retail: product popularity (songs)
Data and Web Mining. - S. Orlando 11
Information Retrieval (IR)
IR helps users find information that matches their
information needs expressed as queries
Historically, IR is about document retrieval,
emphasizing document as the basic unit.
Finding documents relevant to user queries
Technically, IR studies the acquisition, organization,
storage, retrieval, and distribution of information.
Data and Web Mining. - S. Orlando 12
IR architecture
Data and Web Mining. - S. Orlando 13
IR queries
Keyword queries
Boolean queries (using AND, OR, NOT)
Phrase queries
Proximity queries
Full document queries
Natural language questions
Data and Web Mining. - S. Orlando 14
Information retrieval models
An IR model governs how a document and a query
are represented and how the relevance of a
document to a user query is defined
Main models:
Boolean model
Vector space model
Statistical language model
etc
Data and Web Mining. - S. Orlando 15
Boolean model
Each document or query is treated as a bag of words
or terms
Ordering of words is not considered
Given a collection of documents D, let
V = {t1, t2, ..., t|V|}
be the set of distinctive words/terms in the collection.
V is called the vocabulary
A weight wij > 0 is associated with each term ti of
a document dj D
wij = 0/1 (absence/presence)
dj = (w1j, w2j, ..., w|V|j)
Data and Web Mining. - S. Orlando 16
Boolean model (contd)
Query terms are combined logically using the Boolean
operators AND, OR, and NOT.
E.g., ((data AND mining) AND (NOT text))
Retrieval
Given a Boolean query, the system retrieves every
document that makes the query logically true
Exact match
The retrieval results are usually quite poor because
term frequency is not considered.
Data and Web Mining. - S. Orlando 17
Boolean model: an Example
Consider a document space defined by three terms,
i.e., the vocabulary / lexicon:
hardware, software, users
A set of documents is defined as:
A1=(1, 0, 0), A2=(0, 1, 0), A3=(0, 0, 1)
A4=(1, 1, 0), A5=(1, 0, 1), A6=(0, 1, 1)
A7=(1, 1, 1) A8=(1, 0, 1), A9=(0, 1, 1)
If the query is: hardware, software
i.e., (1, 1, 0)
what documents should be retrieved?
AND: documents A4, A7
OR: all documents, but A3
Data and Web Mining. - S. Orlando 18
Similarity matching: an Example
A set of documents is defined as:
A1=(1, 0, 0), A2=(0, 1, 0), A3=(0, 0, 1)
A4=(1, 1, 0), A5=(1, 0, 1), A6=(0, 1, 1)
A7=(1, 1, 1) A8=(1, 0, 1), A9=(0, 1, 1)
In similarity matching (cosine in the Boolean vector
space):
q=(1, 1, 0)
S(q, A1)=0.71, S(q, A2)=0.71, S(q, A3)=0
S(q, A4)=1, S(q, A5)=0.5, S(q, A6)=0.5
S(q, A7)=0.82, S(q, A8)=0.5, S(q, A9)=0.5
Document retrieved set (with ranking, where cosine>0):
{A4, A7, A1, A2, A5, A6, A8, A9}
Data and Web Mining. - S. Orlando 19
Vector space model
Documents are still treated as a bag of words or
terms.
Each document is still represented as a vector.
However, the term weights are not forced to be 0 or 1,
like in the Boolean model
Each term weight is computed on the basis of some variations
of TF or TF-IDF scheme.
Term Frequency (TF) Scheme: The weight of a term ti in
document dj is the number of times that ti appears in dj,
denoted by fij. Normalization may also be applied.
Data and Web Mining. - S. Orlando 20
TF-IDF term weighting scheme
The most well known
weighting scheme
TF: term frequency
IDF: inverse document
frequency.
N: total number of docs
dfi: the number of docs
where ti appears
The final TF-IDF term
weight is:
Data and Web Mining. - S. Orlando 21
Retrieval in vector space model
Query q is represented in the same way or slightly
differently.
Relevance of di to q: Compare the similarity of
query q and document di, i.e. the similarity between
the two associated vectors.
Cosine similarity (the cosine of the angle
between the two normalized vectors)
Cosine is also commonly used in text clustering
Data and Web Mining. - S. Orlando 22
Retrieval in the vector space model
Not only documents dj D, but also queries q are
represented as a vector of weigths of |V| elements:
dj =(w1j, w2j, ..., w|V|j) q =(w1q, w2q, ..., w|V|j)
Data and Web Mining. - S. Orlando 23
23
Retrieval in the vector space model
Rilevance of di w.r.t. q: Compare the similarity of query q
and document di, i.e., the similarity between the two
associated vectors
Cosine similarity (the cosine of the angle)
Example with two
2-dimensional vectors
|V| = 2
24
Data and Web Mining. - S. Orlando 24
Relevance feedback
Relevance feedback is one of the techniques for
improving retrieval effectiveness. The steps:
the user first identifies some relevant (Dr) and irrelevant
documents (Dir) in the initial list of retrieved documents
goal: expand the query vector in order to maximize similarity
with relevant documents, while minimizing similarity with
irrelevant documents
query q expanded by extracting additional terms from the sample
of relevant (Dr) and irrelevant (Dir) documents to produce qe
perform a second round of retrieval
Rocchio method (, and are parameters)
Data and Web Mining. - S. Orlando 25
Rocchio text classifier
Training set: relevant and irrelevant docs
you can train a classifier
The Rocchio classification method, can be used to
improve retrieval effectiveness too
Rocchio classifier is constructed by producing a
prototype vector ci for each class i (relevant or
irrelevant in this case) associated with document
set Di:
Each
vector is
normalized
(sz=1)
In classification, cosine is used
the class is determined by the closest class prototype
(1NN) Data and Web Mining. - S. Orlando 26
Text pre-processing
Document parsing for word (term) extraction: easy
Stopwords removal
Stemming
Frequency counts and computing TF-IDF term
weights.
Data and Web Mining. - S. Orlando 27
Stopwords removal
Many of the most frequently used words in English are useless
in IR and text mining these words are called stop words
the, of, and, to, .
Typically about 400 to 500 such words
For an application, an additional domain specific stopwords
list may be constructed
Why do we need to remove stopwords?
Reduce indexing (or data) file size
stopwords accounts 20-30% of total word counts.
Improve efficiency and effectiveness
stopwords are not useful for searching or text mining
they may also confuse the retrieval system
Current Web Search Engines generally do not use stopword
lists for phrase search queries
Data and Web Mining. - S. Orlando 28
Stemming
Techniques used to find out the root/stem of a
word. e.g.,
user engineering
users engineered
used engineer
using
use stem engineer stem
Usefulness:
improving effectiveness of IR and text mining
Matching similar words
Mainly improve recall
reducing indexing size
combing words with the same roots may reduce indexing
size as much as 40-50%
Web Search Engine may need to index un-stemmed words
too for phrase search
Data and Web Mining. - S. Orlando 29
Basic stemming methods
Using a set of rules. e.g., English rules
remove ending
if a word ends with a consonant other than s,
followed by an s, then delete s.
if a word ends in es, drop the s.
if a word ends in ing, delete the ing unless the remaining
word consists only of one letter or of th.
If a word ends with ed, preceded by a consonant, delete
the ed unless this leaves only a single letter.
...
transform words
if a word ends with ies, but not eies or aies, then
ies y
Data and Web Mining. - S. Orlando 30
Evaluation: Precision and Recall
Given a query:
Are all retrieved documents relevant?
Have all the relevant documents been retrieved?
Measures for system performance:
The first question is about the precision of the search
The second is about the completeness (recall) of the search
By increasing the number of retrieved items, we usually
increase the recall, but also reduce precision
see next slide, where we plot recall vs. precision, obtained by
increasing the size of the result set of a given query Data and Web Mining. - S. Orlando 31
Precision-recall curve
Data and Web Mining. - S. Orlando 32
Compare different retrieval algorithms
Data and Web Mining. - S. Orlando 33
Compare with multiple queries
Compute the average precision at each recall level
Draw precision recall curves
Do not forget the F-score/F-measure evaluation
Data and Web Mining. - S. Orlando 34
Rank precision
Compute the precision values at some selected rank
positions.
Mainly used in Web search evaluation
For a Web search engine, we can compute
precisions for the top 5, 10, 15, 20, 25 and 30
returned pages
as the user seldom looks at more than 30 pages
P@5, P@10, P@15, P@20, P@25, P@30
Recall is not very meaningful in Web search.
Why?
Data and Web Mining. - S. Orlando 35
Inverted index
The inverted index of a document collection is
basically a data structure that
attaches each distinctive term with a list of all
documents that contain the term.
Thus, in retrieval, it takes constant time to
find the documents that contains a query term.
Multiple query terms are also easy handled as we
will see soon.
Data and Web Mining. - S. Orlando 36
An example
DocID, Count,
[position list]
lexicon postings list lexicon postings list
Data and Web Mining. - S. Orlando 37
Index construction
Easy! See the example,
Data and Web Mining. - S. Orlando 38
Index compression
Postings lists are ordered by docIDs
Compression instead of docIDs we can compress smaller gaps
between docIDs, thus reducing space requirements for the index
Use a variable number of bit/byte for gap representation
the gaps have a smaller magnitude than docIDs
apple 1,2,3,5 1,1,1,2
pear 2,4,5 2,2,1
tomato 3,5 3,2 dGap0 = docID0
dGapi>0 = docIDi - docID(i-1)
dGap
Data and Web Mining. - S. Orlando 39
Index compression
Example of compression using Variable Byte econding
7 bits
510= 1012
{
82410= 110 01110002
7 bits
{
21457710= 1101 0001100 01100012
Data and Web Mining. - S. Orlando 40
Search using inverted index
Given a query q, search has the following steps:
Step 1 (Vocabulary search): find each term/word in q in the
inverted index.
Step 2 (Results merging): Merge results to find documents that
contain all or some of the words/terms in q
AND/OR of postings lists
Step 3 (Rank score computation): To rank the resulting
documents/pages, by using
content-based ranking (e.g. TF-IDF)
link-based ranking Web Search Engine
etc. etc.
Data and Web Mining. - S. Orlando 41
Mission impossible ?
WSE
Crawl and index billions of pages
Answer hundreds of millions of queries
per day
In less than 1 sec. per query
Users
Want to submit short queries (on avg.
2.5 terms), often with orthographic errors
Expect to receive the most relevant
results of the Web
In a blink of eye
In terms of 1990 IR, almost unimaginable
42 Data and Web Mining. - S. Orlando 42
Web Search as a huge IR system
Sponsored Links
CG Appliance Express
Discount Appliances (650) 756-3931
User
Same Day Certified Installation
www.cgappliance.com
San Francisco-Oakland-San Jose,
CA
Miele Vacuum Cleaners
Miele Vacuums- Complete Selection
Free Shipping!
www.vacuums.com
Miele Vacuum Cleaners
Miele-Free Air shipping!
All models. Helpful advice.
www.best-vacuum.com
Web Results 1 - 10 of about 7,310,000 for miele. (0.12 seconds)
Miele, Inc -- Anything else is a compromise
At the heart of your home, Appliances by Miele. ... USA. to miele.com. Residential Appliances.
Vacuum Cleaners. Dishwashers. Cooking Appliances. Steam Oven. Coffee System ...
Web spider
www.miele.com/ - 20k - Cached - Similar pages
Miele
Welcome to Miele, the home of the very best appliances and kitchens in the world.
www.miele.co.uk/ - 3k - Cached - Similar pages
Miele - Deutscher Hersteller von Einbaugerten, Hausgerten ... - [ Translate this
page ]
Das Portal zum Thema Essen & Geniessen online unter www.zu-tisch.de. Miele weltweit
...ein Leben lang. ... Whlen Sie die Miele Vertretung Ihres Landes.
www.miele.de/ - 10k - Cached - Similar pages
Herzlich willkommen bei Miele sterreich - [ Translate this page ]
Herzlich willkommen bei Miele sterreich Wenn Sie nicht automatisch
weitergeleitet werden, klicken Sie bitte hier! HAUSHALTSGERTE ...
www.miele.at/ - 3k - Cached - Similar pages
Search
Indexer
The Web
Indexes Ad indexes
Data and Web Mining. - S. Orlando 43
Different search engines
The real differences between different search
engines are
their index weighting schemes
Including context where terms appear, e.g., title, body,
emphasized words, etc.
their query processing methods (e.g., query
classification, expansion, etc)
their ranking algorithms
few of these are published by any of the search
engine companies. They are tightly guarded
secrets.
Data and Web Mining. - S. Orlando 44
Web Search Engines
45 Data and Web Mining. - S. Orlando 45
Web Search Engines: what do the users search?
The 250 most frequent terms in the famous AOL
query log! Data and Web Mining. - S. Orlando 46
Query analysis to evaluate user needs
Informational want to learn about something (~40% / 65%)
Low hemoglobin
Navigational want to go to that page (~25% / 15%)
United Airlines
Transactional want to do something (web-mediated) (~35% /
20%)
Access a service Seattle weather
Downloads Mars surface images
Shop Canon S410
Gray areas
Find a good hub Car rental Brasil
Exploratory search see whats there
A. Z. Broder, A taxonomy of web search, SIGIR Forum, vol. 36, no. 2, pp. 310, 2002.
Data and Web Mining. - S. Orlando 47
Anatomy of a modern Web Search Engine
A. Arasu et al., Searching the Web, ACM Transaction on Internet Technology, 1(1), 2001.
Data and Web Mining. - S. Orlando 48
Crawler
49 Data and Web Mining. - S. Orlando 49
Crawler
It is a program that navigates the Web following the
hyperlinks and stores them in a page repository
Design Issues of the Crawl module:
What pages to download
When to refresh
Minimize load on web sites
How to parallelize the process
Page selection during crawling: Importance metric
Given a page P, define how good that page is, on the basis of
several metrics (combination of them):
Popularity driven: Incoming-link counts (or PageRank)
Location driven: Deepness of the page in a site
Usage driven: Click counts of the pages (feedback)
Interest driven: driven from a query, based on the similarity with
page contents (focused crawling)
Data and Web Mining. - S. Orlando 50
Indexer and Page Repository
51 Data and Web Mining. - S. Orlando 51
Storage: Page repository
The Page Repository is a scalable storage system for web
pages
Allows the Crawler to store pages
Allows the Indexer and Collection Analysis to retrieve them
Similar to other data storage systems DB or file systems
Does not have to provide some of the other systems
features: transactions, logging, directory.
Data and Web Mining. - S. Orlando 52
Designing a Distributed Page Repository
Repository designed to work over a cluster of
interconnected nodes
Page distribution across nodes
Uniform distribution any page can be sent to any node
Hash distribution policy hash page ID space into node ID space
Physical organization within a node
Update strategy
batch (Periodically executed)
steady (Run all the time)
Data and Web Mining. - S. Orlando 53
Indexer and collection analysis modules
The Indexer module creates two indexes:
Text (content) index : Uses Traditional indexing
methods like Inverted Indexing.
Structure (links) index : Uses a directed graph of pages
and links.
Sometimes also creates an inverted graph, in order to
answer queries that ask for all the pages that have
hyperlinks pointing to a given page
The collection analysis module uses the 2 basic indexes
created by the indexer module in order to assemble
Utility Indexes
e.g.: a site index.
Data and Web Mining. - S. Orlando 54
Indexer: Design Issues and Challenges
Index build must be :
Fast
Economic
(unlike traditional index builds)
Incremental indexing must be supported
Personalization
Storage : compression vs. speed
Data and Web Mining. - S. Orlando 55
Index partitioning
Partitioning Inverted Files
Local inverted file
each node contains indexes
of a disjoint partition of the
document collection
query is broadcasted and
answers are obtained by
merging local results
Global inverted file
each node is only responsible
for a subset of terms in the
collection
query is selectively sent to the
appropriate nodes only
Data and Web Mining. - S. Orlando 56
Query engine
57 Data and Web Mining. - S. Orlando 57
Query Engine
Snippet
Decreasing
order of page
importance (ranking)
58 Data and Web Mining. - S. Orlando 58
Query engine
The query engine module accepts queries from multitudes of users
and returns the results
Exploits the partitioned index to quickly find the relevant pages
Use Page Repository to prepare the page of the (10) results
snippet construction is query-based
Since the possible results are a huge number, the ranking
module has to order the results according to their relevance
Ranking
not only based on traditional IR content-based approaches
terms may be of poor quality or not relevant
insufficient self-description of user intent
combat spam
Link analysis, e.g. PageRank that exploits incoming links
from important pages to raise the rank of pages
Exploit proximity of query terms in the pages
Learning to rank
Data and Web Mining. - S. Orlando 59
Summary
We only gave a VERY brief introduction to IR.
There are a large number of other topics, e.g.,
Statistical language model
Latent semantic indexing (LSI and SVD).
Many other interesting topics are not covered, e.g.,
Web search
Index compression
Ranking: combining contents and hyperlinks (see the next
block of slides)
Web page pre-processing
Combining multiple rankings and meta search
Web spamming
Read the textbooks
Data and Web Mining. - S. Orlando 60