Topic-oriented query expansion for web search

ABSTRACT

The contribution of this paper includes three folders: (1) To introduce a topic-oriented query expansion model based on the Information Bottleneck theory that classify terms into distinct topical clusters in order to find out candidate terms for the query expansion. (2) To define a term-term similarity matrix that is available to improve the term ambiguous problem. (3) To propose two measures, intracluster and intercluster similarities, that are based on proximity between the topics represented by two clusters in order to evaluate the retrieval effectiveness. Results of several evaluation experiments in Web search exhibit the average intracluster similarity was improved for the gain of 79.1% while the average intercluster similarity was decreased for the loss of 36.0%.

REFERENCES

• 1: Ricardo A. Baeza-Yates , Berthier Ribeiro-Neto, Modern Information Retrieval, Addison-Wesley Longman Publishing Co., Inc., Boston, MA, 1999
• 2: E.-N. Efthimiadis. Query expansion, 1996.
• 3: M. Sanderson. Word sense disambiguation and information retrieval, 1994.
• 4: N. Slonim et al.. Unsupervised document classification using sequential information maximization, 2002.
• 5: N. Tishby et al. The information bottleneck method, 1999.
• 6: S. Hinrich. Automatic word sense discrimination, 1998.
• 7: T. Hofmann. Probabilistic latent semantic indexing, 1999.
• 8: V. Lavrenko and W. B. Croft. Relevance-based language models, 2001.

INDEX TERMS

Primary Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval
          Subjects: Clustering

Additional Classification:
  H. Information Systems
  H.3 INFORMATION STORAGE AND RETRIEVAL
      H.3.3 Information Search and Retrieval
          Subjects: Query formulation; Retrieval models

General Terms:
Experimentation, Measurement, Theory

Keywords:
information bottleneck, intercluster similarity, intracluster similarity, query expansion, term-term similarity matrix, topic-oriented