Paper: Indexing by Latent Semantic Analysis - PowerPoint PPT Presentation

About This Presentation
Title:

Paper: Indexing by Latent Semantic Analysis

Description:

One term may have multiple meaning. Different terms may have the same meaning. ... c2 A survey of user opinion of computer system response time ... – PowerPoint PPT presentation

Number of Views:44
Avg rating:3.0/5.0
Slides: 14
Provided by: haiya5
Category:

less

Transcript and Presenter's Notes

Title: Paper: Indexing by Latent Semantic Analysis


1
Paper Indexing by Latent Semantic Analysis
  • for course cs630
  • presented by Haiyan Qiao

2
Problem Introduction
  • Traditional term-matching method doesnt work
    well in information retrieval
  • We want to capture the concepts instead of words.
    Concepts are reflected in the words. However,
  • One term may have multiple meaning
  • Different terms may have the same meaning.

3
LSI (Latent Semantic Analysis)
  • LSI approach tries to overcome the deficiencies
    of term-matching retrieval by treating the
    unreliability of observed term-document
    association data as a statistical problem.
  • The goal is to find effective models to represent
    the relationship between terms and documents.
    Hence a set of terms, which is by itself
    incomplete and unreliable, will be replaced by
    some set of entities which are more reliable
    indicants.

4
SVD (Singular Value Decomposition)
  • How to learn the concepts from data?
  • SVD is applied to derive the latent semantic
    structure model.
  • What is SVD?
  • http//kwon3d.com/theory/jkinem/svd.html
  • http//mathworld.wolfram.com/SingularValueDecompo
    sition.html
  • http//www.cs.ut.ee/toomas_l/linalg/lin2/node13.
    htmlSECTION00013200000000000000

5
SVD cont
  • SVD of the term-by-document matrix X
  • If the singular values of S0 are ordered by size,
    we only keep the first k largest values and get a
    reduced model
  • doesnt exactly match X and it gets closer
    as more and more singular values are kept
  • This is what we want. We dont want perfect fit
    since we think some of 0s in X should be 1 and
    vice versa.
  • It reflects the major associative patterns in the
    data, and ignores the smaller, less important
    influence and noise.

6
Fundamental Comparison Quantities from the SVD
Model
  • Comparing Two Terms the dot product between two
    row vectors of reflects the extent to which
    two terms have a similar pattern of occurrence
    across the set of document.
  • Comparing Two Documents dot product between two
    column vectors of
  • Comparing a Term and a Document

7
Example -Technical Memo
  • Query human-computer interaction
  • Dataset
  • c1 Human machine interface for Lab ABC computer
    application
  • c2 A survey of user opinion of computer system
    response time
  • c3 The EPS user interface management system
  • c4 System and human system engineering testing
    of EPS
  • c5 Relations of user-perceived response time to
    error measurement
  • m1 The generation of random, binary, unordered
    trees
  • m2 The intersection graph of paths in trees
  • m3 Graph minors IV Widths of trees and
    well-quasi-ordering
  • m4 Graph minors A survey

8
Example cont
  • 12-term by 9-document matrix
  • gtgt X 1 0 0 1 0 0 0 0 0
  • 1 0 1 0 0 0 0 0 0
  • 1 1 0 0 0 0 0 0 0
  • 0 1 1 0 1 0 0 0 0
  • 0 1 1 2 0 0 0 0 0
  • 0 1 0 0 1 0 0 0 0
  • 0 1 0 0 1 0 0 0 0
  • 0 0 1 1 0 0 0 0 0
  • 0 1 0 0 0 0 0 0 1
  • 0 0 0 0 0 1 1 1 0
  • 0 0 0 0 0 0 1 1 1
  • 0 0 0 0 0 0 0 1 1

9
Example cont
  • XT0S0D0', T0 and D0 have orthonormal columns
    and So is diagonal
  • T0 is the matrix of eigenvectors of the square
    symmetric matrix XX'
  • D0 is the matrix of eigenvectors of XX
  • S0 is the matrix of eigenvalues in both cases
  • gtgt T0, S0 eig(XX')
  • gtgt T0
  •  T0
  •   0.1561 -0.2700 0.1250 -0.4067
    -0.0605 -0.5227 -0.3410 -0.1063 -0.4148
    0.2890 -0.1132 0.2214
  • 0.1516 0.4921 -0.1586 -0.1089
    -0.0099 0.0704 0.4959 0.2818 -0.5522
    0.1350 -0.0721 0.1976
  • -0.3077 -0.2221 0.0336 0.4924
    0.0623 0.3022 -0.2550 -0.1068 -0.5950
    -0.1644 0.0432 0.2405
  • 0.3123 -0.5400 0.2500 0.0123
    -0.0004 -0.0029 0.3848 0.3317 0.0991
    -0.3378 0.0571 0.4036
  • 0.3077 0.2221 -0.0336 0.2707
    0.0343 0.1658 -0.2065 -0.1590 0.3335
    0.3611 -0.1673 0.6445
  • -0.2602 0.5134 0.5307 -0.0539
    -0.0161 -0.2829 -0.1697 0.0803 0.0738
    -0.4260 0.1072 0.2650
  • -0.0521 0.0266 -0.7807 -0.0539
    -0.0161 -0.2829 -0.1697 0.0803 0.0738
    -0.4260 0.1072 0.2650
  • -0.7716 -0.1742 -0.0578 -0.1653
    -0.0190 -0.0330 0.2722 0.1148 0.1881
    0.3303 -0.1413 0.3008
  • 0.0000 0.0000 0.0000 -0.5794
    -0.0363 0.4669 0.0809 -0.5372 -0.0324
    -0.1776 0.2736 0.2059
  • 0.0000 0.0000 0.0000 -0.2254
    0.2546 0.2883 -0.3921 0.5942 0.0248
    0.2311 0.4902 0.0127
  • -0.0000 -0.0000 -0.0000 0.2320
    -0.6811 -0.1596 0.1149 -0.0683 0.0007
    0.2231 0.6228 0.0361
  • 0.0000 -0.0000 0.0000 0.1825
    0.6784 -0.3395 0.2773 -0.3005 -0.0087
    0.1411 0.4505 0.0318

10
Example cont
  • gtgt D0, S0 eig(X'X)
  • gtgt D0
  •  D0
  •    0.0637 0.0144 -0.1773 0.0766
    -0.0457 -0.9498 0.1103 -0.0559 0.1974
  • -0.2428 -0.0493 0.4330 0.2565
    0.2063 -0.0286 -0.4973 0.1656 0.6060
  • -0.0241 -0.0088 0.2369 -0.7244
    -0.3783 0.0416 0.2076 -0.1273 0.4629
  • 0.0842 0.0195 -0.2648 0.3689
    0.2056 0.2677 0.5699 -0.2318 0.5421
  • 0.2624 0.0583 -0.6723 -0.0348
    -0.3272 0.1500 -0.5054 0.1068 0.2795
  • 0.6198 -0.4545 0.3408 0.3002
    -0.3948 0.0151 0.0982 0.1928 0.0038
  • -0.0180 0.7615 0.1522 0.2122
    -0.3495 0.0155 0.1930 0.4379 0.0146
  • -0.5199 -0.4496 -0.2491 -0.0001
    -0.1498 0.0102 0.2529 0.6151 0.0241
  • 0.4535 0.0696 -0.0380 -0.3622
    0.6020 -0.0246 0.0793 0.5299 0.0820

11
Example cont
  • gtgt S0eig(X'X)
  • gtgt S0S0.0.5
  • S0
  •   0.3637
  • 0.5601
  • 0.8459
  • 1.3064
  • 1.5048
  • 1.6445
  • 2.3539
  • 2.5417
  • 3.3409
  • We only keep the largest two singular values
  • and the corresponding columns from the T and D

12
Example cont
  • gtgt T0.2214 -0.1132
  • 0.1976 -0.0721
  • 0.2405 0.0432
  • 0.4036 0.0571
  • 0.6445 -0.1673
  • 0.2650 0.1072
  • 0.2650 0.1072
  • 0.3008 -0.1413
  • 0.2059 0.2736
  • 0.0127 0.4902
  • 0.0361 0.6228
  • 0.0318 0.4505
  • gtgt S 3.3409 0 0 2.5417
  • gtgt D 0.1974 0.6060 0.4629 0.5421
    0.2795 0.0038 0.0146 0.0241 0.0820
  • -0.0559 0.1656 -0.1273 -0.2318
    0.1068 0.1928 0.4379 0.6151 0.5299
  • gtgt TSD
  • 0.1621 0.4006 0.3790 0.4677
    0.1760 -0.0527
  • 0.1406 0.3697 0.3289 0.4004
    0.1649 -0.0328
  • 0.1525 0.5051 0.3580 0.4101 0.2363
    0.0242

13
Summary
  • What is the common and difference between PCA and
    SVD?
  • Both are related to standard eigenvalue-eigenvecto
    r, to remove noise or correlation and get the
    most important info.
  • PCA is on covariance matrix and SVD works on
    original matrix.
Write a Comment
User Comments (0)
About PowerShow.com