8/9/2019 Cluster Selection
1/21
Improving
Web Clustering by
Cluster Selection
By- Vishal Rathore
Regd. No. 0721215022(+91)9861084119
8/9/2019 Cluster Selection
2/21
Web Search
Iterative Process
Problems with Standard Web Search Many Irrelevant Results
Single Long List
Solution Identify and Present Implicit Clusters
2
8/9/2019 Cluster Selection
3/21
Web Clustering
1. Jaguar
Official worldwide web site of Jaguar Cars.
2. Apple - Mac OS XThe Apple Mac OS X product page.
3. Jaguar UK - R is for Racing
The essence of the Jaguar breed
4. Jaguar
General information from Big Cats Online.
5. Jaguar AU - Jaguar Cars
Services and news
6. Jaguar -- Defenders of Wildlife
Size, appearance, life span and diet.
Search Results for: Jaguar 1 6 of 70,000,000
Clusters
1.Car2. Animal
3. Mac OS
4. Other
4. Jaguar
General information from Big Cats Online.
6. Jaguar -- Defenders of WildlifeSize, appearance, life span and diet.
3
8/9/2019 Cluster Selection
4/21
Web Clustering Algorithms
Many standard clustering algorithms.
Text oriented clustering algorithms STC - Suffix Tree Clustering
ESTC - Improvement on STC
4
8/9/2019 Cluster Selection
5/21
Suffix Tree Clustering
Clean Pages
Identify Base Clusters
Combine Base Clusters
Rank/Select Clusters
5
8/9/2019 Cluster Selection
6/21
STC: Identify Base Clusters
car model
car
animal
mac os x
Base clusters each given a score: # documentsv
phrase score
5
10
4.5
24
6
8/9/2019 Cluster Selection
7/21
STC: Combining Base Clusters
3018
Merge Clusters Based On Overlap
Merged ClusterScore is sum of base cluster scores
12
6
7
7
8/9/2019 Cluster Selection
8/21
STC: Rank/Select Clusters
Sort Clusters by Score
Select Best N
8
8/9/2019 Cluster Selection
9/21
Problems with STC
STC is better than many other algorithms
BUT not good enough
Scores
Poor Cluster Quality Measure
Selection Poor Coverage
Excessive Overlap
9
8/9/2019 Cluster Selection
10/21
ESTC: Better Cluster Scoring
Base Cluster Scores OK
Combined Cluster Scores BAD
Overlap between clusters over counted in sum
Example - Particularly Similar Pages
10
8/9/2019 Cluster Selection
11/21
ESTC: Scoring Solution
Solution
Eliminate the over counting of the overlap
Merged Cluster Score
Sum over document scores
Document Score
Average phrase score of base clusterscontaining the document in the merged cluster
11
8/9/2019 Cluster Selection
12/21
ESTC: Better Cluster Selection
Top N Clusters BAD
Dominant Topic over represented
12
Cars Animals Mac OS Other
8/9/2019 Cluster Selection
13/21
ESTC: Smarter Selection The Search
ESTC: Smarter selection
Heuristic
Minimize Overlap
Maximize Coverage
13
8/9/2019 Cluster Selection
14/21
Incremental
Greedy
Look-ahead Protection
Sophisticated Branch and Bound Pruning
ESTC: The Search
14
8/9/2019 Cluster Selection
15/21
Evaluation Method
Gold Standard - Ideal Clustering
2 Searches and 2 Types of Input Data
Jaguar and Salsa
Snippets and Full Text
Precision
Cluster accuracy against the best matching ideal cluster
Recall
Coverage of ideal cluster in matched clusters F-measure
Combination of precision and recall
15
8/9/2019 Cluster Selection
16/21
Results STC, STC-NS, ESTC
Jaguar Full Text Clustering Results
16
8/9/2019 Cluster Selection
17/21
Results ESTC vs Grokker
Similar performance without page titles
Page titles are often very useful
Algorithm Input F-measure
ESTC Snippets 58%
Grokker Snippets + Page Titles 62%ESTC Full Text 74%
17
8/9/2019 Cluster Selection
18/21
Conclusions
ESTC has
A new cluster scoring
A new cluster selection algorithm
ESTC is better than STC, and compares favourablywith Grokker.
ESTC Scoring function applicable to any
agglomerative clustering algorithm.
ESTC Cluster Selection algorithm more widely
applicable.
18
8/9/2019 Cluster Selection
19/21
Future Work
Make improvements to other stages of STC
Particularly Combining Base Clusters
Apply cluster selection method to other
algorithms
Improve cluster selection heuristic
19
8/9/2019 Cluster Selection
20/21
8/9/2019 Cluster Selection
21/21