Incorporating Hierarchical Domain Information to Disambiguate Very Short Queries

Abstract

Users often express their information needs using incomplete or ambiguous queries of only one or two terms in length, particularly in the Web environments. The ambiguity of short queries is a recognized problem for information retrieval (IR) systems. In this study, we investigate various approaches for incorporating hierarchical domain information into IR models such that the domain specification resolves the ambiguity. To this end, we develop practical models for constructing evaluation datasets from existing corpora. In terms of effectiveness, we further study the trade-off between a short query and its domain specification information. In doing so, we find that domains with the highest number of relevant documents are not always the best ones to select. We also evaluate the utility of a domain hierarchy and find that incorporating the hierarchical structure of a collection into the retrieval model could have a high impact on short query disambiguation.