원문정보
초록
영어
Web clustering engine greatly simplifies the effort of the user from browsing the large set of search results by reorganizing them into smaller clusters. Current web clustering engines result in additional clusters and misses out few relevant, leading to lack of predictability of clustering outputs. Web clustering engines produces inconsistent results as the content of the cluster do not always correspond to its label. In this paper, a new web clustering engine named SRCluster has been proposed to overcome these deficiencies, in specific for the polysemy unigram search keywords. SRCluster identifies the possible categories and its label for the given polysemy keyword based on Wikipedia. The system determines the improved Lesk score (termed, SRLesk score) for each of the category. The search result is clustered to the category with the maximum SRLesk score. The hypertext of the disambiguation Wikipedia page is utilized for labeling the cluster. The experimental result on AMBIENT dataset shows that the inconsistency and the lack of predictability of clustering outputs is being improved using SRCluster.
목차
1. Introduction
2. Related Works
2.1 Web Clustering Engines
2.2 Word Sense Disambiguation
2.3 Clustering Methodologies with External Knowledge Resource
2.4 Labeling the Cluster
3. Overview of Web Clustering Engine
4. Knowledge Resource of SRCluster
5. Overlap of Sense Definition
5.1 Traditional Lesk Approaches
5.2 SRLesk : Extended Lesk Approach for SRCluster
6. SRCluster
6.1 Overview
6.2 Architecture
6.3 Result Extractor
6.4 Result Feature Builder
6.5 Concept Identifier & Labeller
6.5 Concept Feature Builder
6.6 SRLesk Clustering Algorithm
6.7 Clustered Search Result
7. Experiments
7.1 Background Information
7.2 Experiments and Results
8. Conclusion
References