test door wim

1. Introduction Clustering Alignment
Doctoral Seminar: Multi-document clustering
and alignment
Wim De Smet
March 23, 2007

2. Introduction Clustering Alignment
Current goals
CLASS, WP7
1. Cluster documents according to topics.
2. Align text and video

3. Introduction Clustering Alignment
Goal
Given news stories about different events, from several sources,
cluster same stories.

4. Introduction Clustering Alignment
Clustering
Typical clustering algorithms: bag of words approach.
Document-by-words matrix:
0.5 0.5 0.5 0 0 0
0.4 0.6 0.5 0 0 0
0.5 0.4 0.6 0 0 0
A= 0 0 0 0.5 0.5 0.5
0 0 0 0.5 0.5 0.5
0.4 0.4 0 0.4 0.4 0.4
0.4 0.4 0.4 0 0.4 0.4

5. Introduction Clustering Alignment
Clustering
Document clustering according to word-similarity:
0.5 0.5 0.5 0 0 0
0.4 0.6 0.5 0 0 0
0.5 0.4 0.6 0 0 0
A= 0 0 0 0.5 0.5 0.5
0 0 0 0.5 0.5 0.5
0.4 0.4 0 0.4 0.4 0.4
0.4 0.4 0.4 0 0.4 0.4

6. Introduction Clustering Alignment
Clustering
Word clustering according to document-similarity:
0.5 0.5 0.5 0 0 0
0.4 0.6 0.5 0 0 0
0.5 0.4 0.6 0 0 0
A= 0 0 0 0.5 0.5 0.5
0 0 0 0.5 0.5 0.5
0.4 0.4 0 0.4 0.4 0.4
0.4 0.4 0.4 0 0.4 0.4

7. Introduction Clustering Alignment
Co-clustering
Purpose: simultaneously clustering words and documents,
preserving information found in both clusterings.
0.5 0.5 0.5 0 0 0
0.4 0.6 0.5 0 0 0
0.5 0.4 0.6 0 0 0
A= 0 0 0 0.5 0.5 0.5
0 0 0 0.5 0.5 0.5
0.4 0.4 0 0.4 0.4 0.4
0.4 0.4 0.4 0 0.4 0.4

8. Introduction Clustering Alignment
Co-clustering
Purpose: simultaneously clustering words and documents,
preserving information found in both clusterings.
0.5 0.5 0.5 0 0 0
0.4 0.6 0.5 0 0 0
0.5 0.4 0.6 0 0 0
A= 0 0 0 0.5 0.5 0.5
0 0 0 0.5 0.5 0.5
0.4 0.4 0 0.4 0.4 0.4
0.4 0.4 0.4 0 0.4 0.4

9. Introduction Clustering Alignment
Co-clustering
Purpose: simultaneously clustering words and documents,
preserving information found in both clusterings.
0.5 0.5 0.5 0 0 0
0.4 0.6 0.5 0 0 0
0.5 0.4 0.6 0 0 0 0.5 0
A= 0 0 0 0.5 0.5 0.5 0 0.5
0 0 0 0.5 0.5 0.5 0.4 0.4
0.4 0.4 0 0.4 0.4 0.4
0.4 0.4 0.4 0 0.4 0.4

10. Introduction Clustering Alignment
Hierarchical Co-clustering
Hierarchical co-clustering:
1. Co-cluster documents and words.
2. For each cluster: if contains too many documents, calculate
sub-matrix
3. Repeat step 1 on sub-matrix.

11. Introduction Clustering Alignment
Bipartite Spectral Graph Partitioning: motivation
View document-by-word matrix as bipartite graph
word1 word2 word2
document1 a1,1 0 0
A=
document2 0 a2,2 a2,3
document2 a3,2 a3,3 0

12. Introduction Clustering Alignment
Bipartite Spectral Graph Partitioning: motivation
Divide graph in document clusters Dm and associated word clusters
Wm ?

13. Introduction Clustering Alignment
Bipartite Spectral Graph Partitioning: motivation
 
 
Wm = wj : Aij ≥ Aij , ∀l = 1, . . . , k
 
i∈Dm i∈Dl

14. Introduction Clustering Alignment
Bipartite Spectral Graph Partitioning: motivation
 
 
Dm = di : Aij ≥ Aij , ∀l = 1, . . . , k
 
j∈Wm j∈Wl

15. Introduction Clustering Alignment
Bipartite Spectral Graph Partitioning: algorithm
1. Given the m ∗ n document-by-word matrix A, calculate
diagonal help-matrices D1 and D2 , so that:
∀1 < i ≤ m : D1 (i, i) = Ai,j
j
∀1 < j ≤ n : D2 (j, j) = Ai,j
i
2. Compute An = D1 −1/2 ∗ A ∗ D2 −1/2
3. Take the SVD of An : SVD(An ) = U ∗ Λ ∗ V∗
4. Determine k, the numbers of clusters by the eigengap:
k = arg max(m≥i>1) λi−1 − λi )/λi−1 , where
λ1 ≥ λ2 ≥ · · · ≥ λm are the singular values of A

16. Introduction Clustering Alignment
Bipartite Spectral Graph Partitioning: algorithm (cont.)
5. From U and V, calculate U[2,··· ,l+1] and V[2,··· ,l+1]
respectively, by taking columns 2 to l + 1
where l = log2 k ,
D1 −1/2 U[2,··· ,l+1]
6. Compute Z = and normalize the rows
D2 −1/2 V[2,··· ,l+1]
of Z
7. Apply k-means to cluster the rows of Z into k clusters
8. Check for each clusters the number of documents. If this is
higher than a given treshold, construct a new
document-by-word matrix formed by the documents and
words in the cluster, and proceed to step 1

17. Introduction Clustering Alignment
Uses of a hierarchical co-clustering
• Documents are clustered according to topic hierarchy
• Words associated with cluster describe topic
• Words can be used for offline clustering

18. Introduction Clustering Alignment
Entries of document-by-word matrix
1. TF-IDF
2. WP 2’s Salience

19. Introduction Clustering Alignment
Results
Precision of clustering 367 news stories from ABC and CNN.
k = defined by eigengap
Salience: 3743 words / TF-IDF: 7242 words
Co-clustering
Test set Precision Recall F1
Salience 74.6 % 41 % 52.9 %
TF-IDF 50.4 % 40.7 % 45.1 %
k-means
Test set Precision Recall F1
Salience 69.5 % 37.1 % 48.4 %
TF-IDF 38.3 % 41.8 % 40 %

20. Introduction Clustering Alignment
Results
Precision of clustering 367 news stories from ABC and CNN.
k = defined by eigengap
Co-clustering
Test set Precision Recall F1
Salience 64.3 % 48.3 % 55.2 %
k-means
Test set Precision Recall F1
Salience 58.3 % 41.7 % 48.8 %

21. Introduction Clustering Alignment
Goals
1. Find aligning segments in
1.1 text-text pairs
1.2 text-video pairs
2. Expand to multiple documents (text and video)

22. Introduction Clustering Alignment
Goals
Using aligned segments:
• Create elaborated story from several sources
• Create links between video and text
• Summarize video and text
• Select appropriate medial form for information

23. Introduction Clustering Alignment
Segments
Segments can be defined at different resolutions
• in text:
• word
• sentence
• paragraph
• in video:
• image
• shot
• Expand to multiple documents (text and video)

24. Introduction Clustering Alignment
Problems
• Degrees of comparability:
• Parallel pairs
• Near-parallel pairs
• Comparable pairs
• Representation of segments in different media: how to
compare

25. Introduction Clustering Alignment
Techniques
• Micro-macro aligment
• Top-down
• Bottom-up
• Make use of several
assumptions:
• Linearity
• Low variance of slope
• Injectivity
• Annealing and Context

26. Introduction Clustering Alignment
Multiple documents
Two possible directions
1. Dimension reduction
2. Expand dimensions of search algorithms