Query Dependent Ranking for Information Retrieval Based on Query Clustering

Ranking is the central problem for information retrieval (IR), and employing machine learning techniques to learn the ranking function is viewed as a promising approach to IR. In information retrieval, the users’ queries often vary a lot from one to another. In this paper we take into account the diversity of query type by clustering the queries. Instead of deriving a single function, this system attempt to develop several ranking functions based on the resulting query clusters in the sense that different queries of the same cluster should have similar characteristics in terms of ranking. Before the queries are clustered, query features are generated based on the average scores of its associated retrieved documents. So, for each query cluster, there will be its associated ranking model. To rank the documents for a new query, the system first find the most suitable cluster for that query and produce the scoring results depend on that cluster. The effectiveness of the system will be tested on LETOR, publicly available benchmark dataset.