Improving Weak Queries using Local Cluster Analysis as a Preliminary Framework

Abstract

In a web retrieval task, the query is usually short and the users expect to find the relevant documents in the first several result pages. To address this issue, the possibilities of using Local Cluster Analysis as a preliminary framework with the intention of improving the effectiveness of weak queries by clustering search results and creating high-precision retrieval is explored in this paper. Moreover, employing this notion makes our approach an apt choice to be embedded in other applications such as Pseudo Relevance Feedback that requires high-precision results and cannot be applied on weak queries currently. The clustering method is notably an important part in our approach. Therefore, the problem of creating effective and meaningful clusters is addressed in this paper and different well-known and state-of-the-art clustering methods are evaluated in order to achieve superior efficiency and effectiveness in the proposed approach. Consequently, various experiments are conducted to evaluate the impact of the proposed architecture and different clustering variants in large Persian text collection created based on TREC specifications. Furthermore, extensive experiments results present promising improvements over existing measures that emphasize on weak queries.