Towards Query Level Resource Weighting for Diversified Query Expansion

Abstract

Diversifying query expansion that leverages multiple resources has demonstrated promising results in the task of search result diversification (SRD) on several benchmark datasets. In existing studies, however, the weight of a resource, or the degree of the contribution of that resource to SRD, is largely ignored. In this work, we present a query level resource weighting method based on a set of features which are integrated into a regression model. Accordingly, we develop an SRD system which generates for a resource a number of expansion candidates that is proportional to the weight of that resource. We thoroughly evaluate our approach on TREC 2009, 2010 and 2011 Web tracks, and show that: 1) our system outperforms the existing methods without resource weighting; and 2) query level resource weighting is superior to the non-query level resource weighting.