Transcript Indexing and vector space

Indexing
The essential step in searching
Review a bit
• We have seen so far
– Crawling
• In the abstract and as implemented
• Your own code and Nutch
• If you are unsure about anything related to
crawling, be sure to speak up now!
– Collection Building
• Once you have crawled, you have a collection of
documents. Presumably, you want to be able to
retrieve the documents that are relevant to
specified information need.
Information Retrieval
• Finding the specific bit of information in
the collection that satisfies a need and
allows a user to complete a task.
• Remember – a web search does not
search the web directly.
– It searches in the index created when the
web pages were found and analyzed.
• Last class, we saw the basic structure of
indexing. Review 
Inverted index construction
Documents to
be indexed.
Token stream.
Modified
tokens.
Friends, Romans, countrymen.
Tokenizer
Romans
Countrymen
friend
roman
countryman
Linguistic modules
Stop words, stemming,
capitalization, cases, etc.
Inverted index.
Friends
Indexer
friend
2
4
roman
1
2
countryman
13
16
Indexer steps: Token
sequence
 Sequence of (Modified token, Document
ID) pairs.
Initially, all the tokens from document 1, then
all the tokens from document 2, etc., without
regard for duplication.
Doc 1
I did enact Julius
Caesar I was killed
i' the Capitol;
Brutus killed me.
Doc 2
So let it be with
Caesar. The noble
Brutus hath told you
Caesar was ambitious
Indexer steps: Sort
 Sort by terms
 docID within terms
Core indexing step
Indexer steps: Dictionary & Postings
 Multiple
term entries
in a single
document
are merged.
 Split into
Dictionary
and Postings
 Doc.
frequency
information
is added.
Number of documents in which the term appears
Spot check
• Complete the indexing for the following two
“documents.”
– Of course, the examples have to be very small to be
manageable. Imagine that you are indexing the entire
news stories.
– Construct the charts as seen on the previous slide
– Put your solution in the Blackboard Indexing – Spot
Check 1. There is a discussion board in Blackboard.
You will find it on the content homepage.
Document 1:
Pearson and Google Jump
Into Learning
Management With a
New, Free System
Document 2:
Pearson adds free learning
management tools to Google
Apps for Education
Problem with Boolean search:
feast or famine
• Boolean queries often result in either too
few (=0) or too many (1000s) results.
• A query that is too broad yields hundreds
of thousands of hits
• A query that is too narrow may yield no hits
• It takes a lot of skill to come up with a
query that produces a manageable number
of hits.
– AND gives too few; OR gives too many
Ranked retrieval models
• Rather than a set of documents satisfying a
query expression, in ranked retrieval
models, the system returns an ordering
over the (top) documents in the collection
with respect to a query
• Free text queries: Rather than a query
language of operators and expressions, the
user’s query is just one or more words in a
human language
• In principle, these are different options, but
in practice, ranked retrieval models have
normally been associated with free text
queries and vice versa
10
Feast or famine: not a problem in
ranked retrieval
• When a system produces a ranked result
set, large result sets are not an issue
– Indeed, the size of the result set is not an issue
– We just show the top k ( ≈ 10) results
– We don’t overwhelm the user
– Premise: the ranking algorithm works
Scoring as the basis of ranked
retrieval
• We wish to return, in order, the documents
most likely to be useful to the searcher
• How can we rank-order the documents in
the collection with respect to a query?
• Assign a score – say in [0, 1] – to each
document
• This score measures how well document
and query “match”.
Query-document matching scores
• We need a way of assigning a score to a
query/document pair
• Let’s start with a one-term query
• If the query term does not occur in the
document: score should be 0
• The more frequent the query term in the
document, the higher the score (should be)
• We will look at a number of alternatives for
this.
Take 1: Jaccard coefficient
• A commonly used measure of overlap of
two sets A and B
– jaccard(A,B) = |A ∩ B| / |A ∪ B|
– jaccard(A,A) = 1
– jaccard(A,B) = 0 if A ∩ B = 0
• A and B don’t have to be the same size.
• Always assigns a number between 0 and 1.
Jaccard coefficient: Scoring
example
• What is the query-document match score
that the Jaccard coefficient computes for
each of the two “documents” below?
• Query: ides of march
• Document 1: caesar died in march
• Document 2: the long march
Jaccard Example done
•
•
•
•
•
Query: ides of march
Document 1: caesar died in march
Document 2: the long march
A = {ides, of, march}
B1 = {caesar, died, in, march}
AI B1  {m arch}
AUB1  {caesar,died,ides,in,of ,m arch}
AI B1/ AUB1  1/6
• B2 = {the, long, march}

AI B2  {m arch}
AUB2  {ides,long,m arch,of,the}
AI B2 / AUB2 1/5
Issues with Jaccard for scoring
• It doesn’t consider term frequency (how
many times a term occurs in a document)
– Rare terms in a collection are more informative
than frequent terms. Jaccard doesn’t consider
this information
• We need a more sophisticated way of
normalizing for length
The problem with the first example was that document 2
“won” because it was shorter, not because it was a better
match. We need a way to take into account document length
so that longer documents are not penalized in calculating the
match score.
Term frequency
• Two factors:
– A term that appears just once in a document is
probably not as significant as a term that
appears a number of times in the document.
– A term that is common to every document in a
collection is not a very good choice for
choosing one document over another to
address an information need.
• Let’s see how we can use these two ideas
to refine our indexing and our search
Term frequency - tf
• The term frequency tft,d of term t in document
d is defined as the number of times that t
occurs in d.
• We want to use tf when computing querydocument match scores. But how?
• Raw term frequency is not what we want:
– A document with 10 occurrences of the term is
more relevant than a document with 1 occurrence
of the term.
– But not 10 times more relevant.
• Relevance does not increase proportionally
with term frequency.
NB: frequency = count in IR
Log-frequency weighting
• The log frequency weight of term t in d is
wt,d
1  log10 t ft,d ,

0,

if t ft,d  0
ot herwise
• 0 → 0, 1 → 1, 2 → 1.3, 10 → 2, 1000 → 4,
etc.
• Score for a document-query pair: sum over
terms t in both q and d:
score  tqd (1  log tft ,d )
• The score is 0 if none of the query terms is
present in the document.
Document frequency
• Rare terms are more informative than
frequent terms
– Recall stop words
• Consider a term in the query that is rare in
the collection (e.g., arachnocentric)
• A document containing this term is very
likely to be relevant to the query
arachnocentric
• → We want a high weight for rare terms like
arachnocentric, even if the term does not
appear many times in the document.
Document frequency, continued
• Frequent terms are less informative than rare terms
• Consider a query term that is frequent in the collection
(e.g., high, increase, line)
– A document containing such a term is more likely to be
relevant than a document that doesn’t
– But it’s not a sure indicator of relevance.
• → For frequent terms, we want high positive weights
for words like high, increase, and line
– But lower weights than for rare terms.
• We will use document frequency (df) to capture this.
idf weight
• dft is the document frequency of t: the
number of documents that contain t
– dft is an inverse measure of the
informativeness of t
– dft  N (the number of documents)
• We define the idf (inverse document
frequency) of t by
idf t  log10 ( N/df t )
– We use log (N/dft) instead of N/dft to
“dampen” the effect of idf.
Will turn out the base of the log is immaterial.
idf example, suppose N = 1 million
term
dft
calpurnia
idft
1
6
animal
100
4
sunday
1,000
3
10,000
2
100,000
1
1,000,000
0
fly
under
the
idf t  log10 ( N/df t )
There is one idf value for each term t in a collection.
Effect of idf on ranking
• Does idf have an effect on ranking for oneterm queries, like
– iPhone
• idf has no effect on ranking one term
queries
– idf affects the ranking of documents for queries
with at least two terms
– For the query capricious person, idf weighting
makes occurrences of capricious count for
much more in the final document ranking than
occurrences of person.
25
Collection vs. Document
frequency
• The collection frequency of t is the number of
occurrences of t in the collection, counting
multiple occurrences.
• Example:
Word
Collection
frequency
Document frequency
insurance
10440
3997
try
10422
8760
• Which word is a better search term (and
should get a higher weight)?
tf-idf weighting
• The tf-idf weight of a term is the product of its
tf weight and its idf weight.
w t ,d  (1  logtft ,d )  log10 ( N / df t )
• Best known weighting scheme in information
retrieval
– Note: the “-” in tf-idf is a hyphen, not a minus sign!
– Alternative names: tf.idf, tf x idf
• Increases with the number of occurrences
within a document
• Increases with the rarity of the term in the
collection
Final ranking of documents for a
query
Score (q,d)  
t qd
tf.idf t,d
28
Binary → count → weight matrix
Antony and Cleopatra
Julius Caesar
The Tempest
Hamlet
Othello
Macbeth
Antony
5.25
3.18
0
0
0
0.35
Brutus
1.21
6.1
0
1
0
0
Caesar
8.59
2.54
0
1.51
0.25
0
Calpurnia
0
1.54
0
0
0
0
Cleopatra
2.85
0
0
0
0
0
mercy
1.51
0
1.9
0.12
5.25
0.88
worser
1.37
0
0.11
4.15
0.25
1.95
Each document is now represented by a real-valued vector of tf-idf
weights ∈ R|V|
|V| is the as
number
Documents
vectors
of terms
• So we have a |V|-dimensional vector space
• Terms are axes of the space
• Documents are points or vectors in this
space
• Very high-dimensional: tens of millions of
dimensions when you apply this to a web
search engine
• These are very sparse vectors - most entries
are zero.
Queries as vectors
• Key idea 1: Do the same for queries:
represent them as vectors in the space
• Key idea 2: Rank documents according to
their proximity to the query in this space
• proximity = similarity of vectors
• proximity ≈ inverse of distance
• Recall: We do this because we want to get
away from the you’re-either-in-or-out
Boolean model.
• Instead: rank more relevant documents
higher than less relevant documents
Making it clear
• We now see a document as a vector: a
sequence of real numbers, each
associated with a term that may or may
not be in the document.
• Every indexed term in the document
collection is represented in the vector.
Usually, that means a lot of vector
entries.
Exercise
• Open the file Search-jaguar.txt
• Skip the first line, which is just a description
of the page.
• Treat each other entry on the page (1 to 3
lines each) as a document.
– Create the dictionary (complete list of terms
for all the collection. Do not eliminate any stop
words, etc.) Do merge singular and plural
forms.
– For each document, create its vector.
• First, just do 1 or 0 – word is in the document or not.
• Then compute the tf-idf weights.
– It is ok to divide this up – each student create the tf-idf
vector for a different document.
What is a vector?
• Suppose we have only two terms in our
dictionary. Suppose jaguar and car.
• Suppose that two documents have the
following tf-idf weights of those two
terms:
– Doc 1: jaguar 0.800
– Doc 2: jaguar 0.300
car 0.600
car 0.500
• How similar are those documents?
Vector space
• Each document is represented by the
sequence of numbers that show its
weight in this two dimensional space:
– Doc 1 = (0.8, 0.6)
– Doc 2 = (0.3, 0.5)
car
We will see later how to normalize so
that all weights are between 0 and 1
.6
.5 Doc 2
Doc 1
.3
.8
jaguar
With this representation, we can
visualize (for 2 dimensions) how
similar two documents are.
More importantly, we can
measure their similarity and
compare it to the similarity of
another document to one of
these.
Dimensionality
• We are showing only two dimensions –
appropriate if there are only two terms.
• There are thousands of terms, perhaps
millions, in the general case.
• We can apply the same measurement
techniques, but we cannot draw them.
• So, how shall we measure the distance
between two vectors?
Formalizing vector space
proximity
• First cut: distance between two points
( = distance between the end points of the
two vectors)
• Euclidean distance?
• Euclidean distance is a bad idea . . .
• . . . because Euclidean distance is large
for vectors of different lengths.
Why distance is a bad idea
The Euclidean
distance between q
and d2 is large even
though the
distribution of terms
in the query q and
the distribution of
terms in the
document d2 are
very similar.
Note that if d2 had fewer references to each of the terms, such that the
vector length was similar to the others, it would be clearly the closest to q.
Is this the result that we want?
Use angle instead of distance
• Thought experiment: take a document d and
append it to itself. Call this document d′
• “Semantically” d and d′ have the same content
• The Euclidean distance between the two
documents can be quite large
• The angle between the two documents is 0,
corresponding to maximal similarity.
• Key idea: Rank documents according to angle
with query.
From angles to cosines
• The following two notions are
equivalent.
– Rank documents in decreasing order of the
angle between query and document
– Rank documents in increasing order of
cosine(query,document)
• Cosine is a monotonically decreasing
function for the interval [0o, 180o]
From angles to cosines
• But how – and why – should we be computing cosines?
Length normalization
• A vector can be (length-) normalized by dividing
each of its components by its length – for this we
use the L2 norm:

x2
2
x
i i
• Dividing a vector by its L2 norm makes it a unit
(length) vector (on surface of unit hypersphere)
• Effect on the two documents d and d′ (d appended
to itself) from earlier slide: they have identical
vectors after length-normalization.
– Long and short documents now have comparable
weights
cosine(query,document)
Dot product
Recall: tf gives significance to terms
r
r
that appear frequently;
idf
gives
r r that are
qd
significance to terms
cos(q, d )  r r
unusual over all the documents
Unit vectors
r r
q d
 r  r 
q d
qd
 -Reminder?

V
qd
i1 i i

V
2
i1 i
q

V
i1
d
2
i
qi is the tf-idf weight of term i in the query
di is the tf-idf weight of term i in the document
cos(q,d) is the cosine similarity of q and d … or,
equivalently, the cosine of the angle between q and d.
In mathematics, the dot product is an algebraic operation that takes two equal-length sequences of numbers
(usually coordinate vectors) and returns a single number obtained by multiplying corresponding entries and
then summing those products. The name is derived from the centered dot " " that is often used to designate
this operation; the alternative name scalar product emphasizes the scalar (rather than vector) nature of the
result. At a basic level, the dot product is used to obtain the cosine of the angle between two vectors.
Cosine for length-normalized
vectors
• For length-normalized vectors, cosine
similarity is simply the dot product (or
scalar product):
cos(q,d )  q  d   qi di
V
i1
for q, d length-normalized.
44
Cosine similarity illustrated
45
Cosine similarity amongst 3 documents
How similar are
the novels
SaS: Sense and
Sensibility
PaP: Pride and
Prejudice, and
WH: Wuthering
Heights?
term
affection
SaS
PaP
WH
115
58
20
jealous
10
7
11
gossip
2
0
6
wuthering
0
0
38
Term frequencies (counts)
Note: To simplify this example, we don’t do idf weighting.
3 documents example contd.
Log frequency weighting
term
SaS
PaP
After length normalization
WH
term
SaS
PaP
WH
affection
3.06
2.76
2.30
affection
0.789
0.832
0.524
jealous
2.00
1.85
2.04
jealous
0.515
0.555
0.465
gossip
1.30
0
1.78
gossip
0.335
0
0.405
0
0
2.58
wuthering
0
0
0.588
wuthering
cos(SaS,PaP) ≈
0.789 × 0.832 + 0.515 × 0.555 + 0.335 × 0.0 + 0.0 × 0.0
≈ 0.94
Recall: log frequency weight is
cos(SaS,WH) ≈ 0.79
if t f  0
1  log t f ,
w


cos(PaP,WH) ≈ 0.69
0,
ot herwise

10
t,d
Why do we have cos(SaS,PaP) > cos(SAS,WH)?
t,d
t,d
Weighting may differ in queries vs
documents
• Many search engines allow for different weightings for queries
vs. documents
• SMART Notation: denotes the combination in use in an engine,
with the notation ddd.qqq, using the acronyms from this table
A very standard combination is lnc.ltc
Table from Wikipedia
tf-idf example: lnc.ltc
Document: car insurance auto insurance
Query: best car insurance
Term
Query
tf- tf-wt
raw
df
idf
Document
wt
n’liz
e
tf-raw
tf-wt
Pro
d
n’liz
e
wt
auto
0
0
5000
2.3
0
0
1
1
1
0.52
0
best
1
1 50000
1.3
1.3
0.34
0
0
0
0
0
car
1
1 10000
2.0
2.0
0.52
1
1
1
0.52
0.27
insurance
1
1
3.0
3.0
0.78
2
1.3
1.3
0.68
0.53
1000
Doc length =
12  02 12 1.32 1.92
Score = 0+0+0.27+0.53 = 0.8
Summary – vector space ranking
• Represent the query as a weighted tf-idf
vector
• Represent each document as a weighted
tf-idf vector
• Compute the cosine similarity score for
the query vector and each document
vector
• Rank documents with respect to the
query by score
• Return the top K (e.g., K = 10) to the user
The Practical Reality
• No, you don’t have to write code from
scratch to do the indexing and searching
functions.
• Just as we found Nutch to do crawling, we
can find professionally developed open
source products for indexing and searching.
• Let’s look briefly at Lucene and Solr.
– Next week – hands on laboratory for
incorporating these into your projects
Lucene – A search resource
• Part of the Apache suite of web-related
software.
• Apache Lucene is a high-performance, fullfeatured text search engine library written
entirely in Java. It is a technology suitable
for nearly any application that requires fulltext search, especially cross-platform.
• Apache Lucene is an open source project
available for free download.
Lucene Features
• Scalable, High-Performance Indexing
–
–
–
–
over 20MB/minute on Pentium M 1.5GHz
small RAM requirements -- only 1MB heap
incremental indexing as fast as batch indexing
index size roughly 20-30% the size of text indexed
• Powerful, Accurate and Efficient Search Algorithms
– ranked searching -- best results returned first
– many powerful query types: phrase queries, wildcard
queries, proximity queries, range queries and more
– fielded searching (e.g., title, author, contents)
– date-range searching
– sorting by any field
– multiple-index searching with merged results
– allows simultaneous update and searching
Lucene overview
• Lucene provides
– Indexing of your information
– Search of your information, using the
previously constructed index.
• You provide
– The user interface
– The application of which the search is a part
Step by step - Index
• First step - Index
– Build an index of whatever you will want to search
later.
• Several classes (Java classes, that is) of analyzers.
Choose the one that suits your need:
– StandardAnalyzer
» A sophisticated general-purpose analyzer.
– WhitespaceAnalyze
» A very simple analyzer that just separates tokens using
white space.
– StopAnalyzer
» Removes common English words that are not usually
useful for indexing.
– SnowballAnalyzer
» An interesting experimental analyzer that works on
word roots (a search on rain should also return entries
with raining, rained, and so on).
Second step - Search
• After the index is built, we can search.
• There are java classes provided (IndexSearcher
and QueryParser) to do just what the names
suggest.
• You specify the analyzer that you want to
apply to the query – it must be the same
analyzer that you used to build the index, not
surprisingly.
• You also specify the field that you want to
search, and the query that the user entered,
that says what you want to search for.
Lucene
• Modified Vector Space Model of search
– combines Boolean and VSM
• Written in Java, but ported to many
other languages
• A library for enabling text-based search
• Note – you must build the application
– Lucene is a collection of modules to use in
your application to do the indexing and
searching steps
Lucene indexes strings
• Not aware of markup, such as XML,
HTML, Word, PDF, etc
• There are good open source tools for
extracting the content from marked up
files
– See BeautifulSoup for HTML, for example
Lucene Analysis
• Analysis is the process of preparing text
for use in indexing and searching
• Analyzer, Tokenizer, TokenFilter classes
– TokenFilter can apply stemming, for
example.
– Several analyzers come with Lucene.
Application developer chooses which to
use.
• Example: WhitespaceAnalyzer separates tokens
based on white space
Lucene - final
• Lucene is a collection of java classes that
implement the type of indexing and
searching techniques we talk about, and
that you read elsewhere.
• The implementation is free, but has to
be integrated into whatever application
you are building to which you want to
add search capability.
Solr
• Built on lucene, solr is an open source
search platform, also from the apache
project.
• Features include “powerful full-text search,
hit highlighting, faceted search, dynamic
clustering, database integration, and rich
document (e.g., Word, PDF) handling. Solr
is highly scalable, providing distributed
search and index replication, and it powers
the search and navigation features of many
of the world's largest internet sites. “
Resources for this material
• Introduction to Information Retrieval, sections
6.2 – 6.4.3
• Slides provided by the author
• http://www.miislita.com/informationretrieval-tutorial/cosine-similaritytutorial.html -- Term weighting and cosine
similarity tutorial
• Better Search with Apache Lucene and Solr
trijug.org/downloads/TriJug-11-07.pdf