CS276A

Text Retrieval and Mining

Lecture 15

Thanks to Thomas Hoffman, Brown University

for sharing many of these slides.

Recap: Clustering 2

Hierarchical clustering

Agglomerative clustering techniques

Evaluation
Term vs. document space clustering
Multi-lingual docs
Feature selection
Labeling

Linear Algebra
Background

Eigenvalues & Eigenvectors

Eigenvectors (for a square mm matrix S)

Example

(right) eigenvector

eigenvalue

How many eigenvalues are there at most?

only has a non-zero solution if
this is a m-th order equation in which can have at
most m distinct solutions (roots of the characteristic
polynomial) can be complex even though S is real.

Matrix-vector multiplication
3 0 0
S 0 2 0
0 0 0

has eigenvalues 3, 2, 0 with

corresponding eigenvectors

1

v1 0
0

0

v2 1
0

0

v3 0
1

On each eigenvector, S acts as a multiple of the

identity
2
matrix: but as a different
multiple on each.

4
Any vector (say x= ) can be viewed as a combination of
6
the eigenvectors:
x = 2v1 + 4v2 + 6v3

Matrix vector multiplication

Thus a matrix-vector multiplication such as Sx (S,

x as in the previous slide) can be rewritten in
terms of the eigenvalues/vectors:

Sx S (2v1 4v2 6v 3 )
Sx 2 Sv1 4 Sv2 6 Sv 3 21v1 42 v2 6 3v 3

Even though x is an arbitrary vector, the action of

S on x is determined by the eigenvalues/vectors.
Suggestion: the effect of small eigenvalues is
small.

Eigenvalues & Eigenvectors

For symmetric matrices, eigenvectors for distinct
eigenvalues are orthogonal

Sv{1, 2} {1, 2}v{1, 2} , and 1 2 v1 v2 0

All eigenvalues of a real symmetric matrix are real.

for complex , if S I 0 and S ST

All eigenvalues of a positive semidefinite matrix
are non-negative

w n , wT Sw 0, then if Sv v 0

Example
2 1
S

1 2

Let

Then

The eigenvalues are 1 and 3 (nonnegative, real).

The eigenvectors are orthogonal (and real):

Real, symmetric.

2
S I
1

1

1

1
2
( 2 ) 1 0.

Plug in these values

and solve for
eigenvectors.

Eigen/diagonal Decomposition

Let
be a square matrix with mlinearly
independent eigenvectors (a non-defective
Uniqu
matrix)
e for
Theorem: Exists an eigen decomposition
diagonal

(cf. matrix diagonalization theorem)

Columns of U are eigenvectors of S

Diagonal elements of

are eigenvalues of

distinc
t
eigenvalues

Diagonal decomposition: why/how

Let U have the eigenvectors as columns: U v1 ... vn

Then, SU can be written

SU S v1 ... vn 1v1 ... n vn v1

Thus SU=U , or U1SU=

And S=U U1.

... vn
...

Diagonal decomposition - example

Recall

2 1
S
; 1 1, 2 3.

1 2

1 and 1
The eigenvectors

1

1
Inverting, we haveU

1 / 2 1 / 2

1
/
2
1
/
2

1 1
formU

1
1

Recall
UU1 =1.

1 1 1 0 1 / 2 1 / 2

Then, S=U U1 =

1 1 0 3 1 / 2 1 / 2

Example continued
Lets divide U (and multiply U1) by 2

1 / 2 1 / 2 1 0 1 / 2
Then, S=

1 / 2 1 / 2 0 3 1 / 2
Q

Why? Stay tuned

1/ 2

1/ 2

(Q-1= QT )

Symmetric Eigen Decomposition

If

is a symmetric matrix:

Theorem: Exists a (unique) eigen

T
decomposition S QQ

where Q is orthogonal:

Q-1= QT

Columns of Q are normalized eigenvectors

Columns are orthogonal.

(everything is real)

Exercise

Examine the symmetric eigen decomposition, if

any, for each of the following matrices:

0 1
1 0

0 1
1 0

1 2
2 3

2 2
2 4

Time out!

I came to this class to learn about text retrieval

and mining, not have my linear algebra past
dredged up again

But if you want to dredge, Strangs Applied

Mathematics is a good place to start.

What do these matrices have to do with text?

Recall m n term-document matrices
But everything so far needs square matrices so

Singular Value Decomposition

For an m n matrix Aofrankrthere exists a factorization
(Singular Value Decomposition = SVD) as follows:

A UV

mm mn

V is nn

The columns of U are orthogonal eigenvectors of AAT.

The columns of V are orthogonal eigenvectors of ATA.
Eigenvalues 1 r of AAT are the eigenvalues of ATA.
i i

diag 1... r

Singular values.

Singular Value Decomposition

Illustration of SVD dimensions and sparseness

SVD example
Let

1 1

A 0 1
1 0

Thus m=3, n=2. Its SVD is

0
1/ 2
1/ 2

2/ 6
1/ 6
1/ 6

1/ 3 1 0
1/ 2

1/ 3 0
3
1/ 2

1 / 3 0 0

1/ 2

1/ 2

Typically, the singular values arranged in decreasing order.

Low-rank Approximation

SVD can be used to compute optimal low-rank

approximations.
Approximation problem: Find Ak of rank k such that

Ak

min

A X

X :rank ( X ) k

Ak and X are both mn matrices.

Typically, want k << r.

Frobenius norm

Low-rank Approximation

Solution via SVD

Ak U diag( 1 ,..., k ,0,...,0)V T

set smallest r-k
singular values to zero

Ak i 1 i ui viT
k

column notation: sum

of rank 1 matrices

Approximation error

How good (bad) is this approximation?

Its the best possible, measured by the Frobenius
norm of the error:

min

X :rank ( X ) k

A X

A Ak

k 1

where the i are ordered such that i i+1.

Suggests why Frobenius error drops as k
increased.

Recall random projection

Completely different method for low-rank

approximation
Was data-oblivious

Error for random projection depended only on

start/finish dimensionality

SVD-based approximation is data-dependent

For every distance

Error for SVD-based approximation is for the

Frobenius norm, not for individual distances

SVD Low-rank approximation

Whereas the term-doc matrix A may have

m=50000, n=10 million (and rank close to 50000)
We can construct an approximation A100 with rank
100.

Of all rank 100 matrices, it would have the lowest

Frobenius error.

Great but why would we??

Answer: Latent Semantic Indexing

C. Eckart, G. Young, The approximation of a matrix by another of lower rank.

Psychometrika, 1, 211-218, 1936.

Latent Semantic
Analysis via SVD

What it is

From term-doc matrix A, we compute the

approximation Ak.

There is a row for each term and a column for

each doc in Ak

Thus docs live in a space of k<<r dimensions

These dimensions are not the original axes

But why?

Vector Space Model: Pros

Automatic selection of index terms

Partial matching of queries and documents

(dealing with the case where no document contains all

search terms)
Ranking according to similarity score (dealing
with large result sets)
Term weighting schemes (improves retrieval
performance)

Various extensions

Document clustering
Relevance feedback (modifying query vector)

Geometric foundation

Problems with Lexical Semantics

Ambiguity and association in natural language

Polysemy: Words often have a multitude of meanings and

different types of usage (more urgent for very
heterogeneous collections).
The vector space model is unable to discriminate between
different meanings of the same word.

Synonymy: Different terms may have an identical or a

similar meaning (weaker: words indicating the same topic).
No associations between words are made in the vector
space representation.

Polysemy and Context

Document similarity on single word level:

polysemy and context

ring
jupiter

meaning 1

planet
...

space
voyager

saturn

...

meaning 2
contribution to similarity, if
used in 1st meaning, but not
if in 2nd

car
company

dodge
ford

Latent Semantic Analysis

Perform a low-rank approximation of document-term

matrix (typical rank 100-300)
General idea

Map documents (and terms) to a low-dimensional

representation.
Design a mapping such that the low-dimensional space
reflects semantic associations (latent semantic space).
Compute document similarity based on the inner product in
this latent semantic space

Goals

Similar terms map to similar location in low dimensional

space
Noise reduction by dimension reduction

Latent Semantic Analysis

Latent semantic space: illustrating example

courtesy of Susan Dumais

Performing the maps

Each row and column of A gets mapped into the

k-dimensional LSI space, by the SVD.
Claim this is not only the mapping with the best
(Frobenius error) approximation to A, but in fact
improves retrieval.
A query q is also mapped into this space, by

qk q U k
T

1
k

Query NOT a sparse vector.

Empirical evidence

Experiments on TREC 1/2/3 Dumais

Lanczos SVD code (available on netlib) due to
Berry used in these expts

Dimensions various values 250-350 reported

Running times of ~ one day on tens of thousands

of docs
(Under 200 reported unsatisfactory)

Generally expect recall to improve what about

precision?

Empirical evidence

Precision at or above median TREC precision

Top scorer on almost 20% of TREC topics

Slightly better on average than straight vector

spaces
Effect of dimensionality:
Dimensions

Precision

250

0.367

300

0.371

346

0.374

Failure modes

Negated phrases

Boolean queries

TREC topics sometimes negate certain

query/terms phrases automatic conversion of
topics to
As usual, freetext/vector space syntax of LSI
queries precludes (say) Find any doc having to do
with the following 5 companies

See Dumais for more.

But why is this clustering?

Weve talked about docs, queries, retrieval and

precision here.
What does this have to do with clustering?
Intuition: Dimension reduction through LSI brings
together related axes in the vector space.

Intuition from block matrices

n documents
Block 1

Whats the rank of this matrix?

0s

Block 2

m
terms

0s
Block k
= non-zero entries.

Intuition from block matrices

n documents
Block 1

0s

Block 2

m
terms

0s
Block k

Vocabulary partitioned into k topics

(clusters); each doc discusses only one topic.

Intuition from block matrices

n documents
Block 1

Whats the best rank-k

approximation to this matrix?
0s

Block 2

m
terms

0s
Block k
= non-zero entries.

Intuition from block matrices

Likely theres a good rank-k
approximation to this matrix.
wiper
tire
V6

Block 1

Block 2

Few nonzero entries

Few nonzero entries

car
10
automobile 0 1

Block k

Simplistic picture
Topic 1
Topic 2

Topic 3

Some wild extrapolation

The dimensionality of a corpus is the number of

distinct topics represented in it.
More mathematical wild extrapolation:

if A has a rank k approximation of low Frobenius

error, then there are no more than k distinct topics
in the corpus.

LSI has many other applications

In many settings in pattern recognition and

retrieval, we have a feature-object matrix.

This matrix may be redundant in dimensionality.

For text, the terms are features and the docs are
objects.
Could be opinions and users more in 276B.
Can work with low-rank approximation.
If entries are missing (e.g., users opinions), can
recover if dimensionality is low.

Powerful general analytical technique

Close, principled analog to clustering methods.

Resources

http://www.cs.utk.edu/~berry/lsi++/
http://lsi.argreenhouse.com/lsi/LSIpapers.html
Dumais (1993) LSI meets TREC: A status report.
Dumais (1994) Latent Semantic Indexing (LSI) and
TREC-2.
Dumais (1995) Using LSI for information filtering:
TREC-3 experiments.
M. Berry, S. Dumais and G. O'Brien. Using linear
algebra for intelligent information retrieval. SIAM
Review, 37(4):573--595, 1995.