Dynamic Clustering of Web Search Results

Abstract

A problem in Web searches is how to help users quickly find useful links from a long list of returned URLs. Document clustering provides an approach to organize retrieval results by clustering documents into meaningful groups. Because a word in a document is naturally correlated with neighboring words, document clustering often uses phrases rather than individual words in determining clusters. We have designed a system to cluster Web search results based on phrases that contain one or more search keywords. We show that, rather than clustering based on whole documents, clustering based on phrases containing search keywords often gives more accurate and informative clusters. Algorithms and experimental results are discussed.