A Case For Shorter Queries, and Helping Users Create Them

Proceedings of NAACL HLT 2007, pages 220–227, Rochester, NY, April 2007. c⃝2007 Association for Computational Linguistics A Case for Shorter Queries, and Helping Users Create Them Giridhar Kumaran and James Allan Center for Intelligent Information Retrieval Department of Computer Science University of Massachusetts Amherst Amherst, MA 01003, USA {giridhar,allan}@cs.umass.edu Abstract Information retrieval systems are fre- quently required to handle long queries. Simply using all terms in the query or re- lying on the underlying retrieval model to appropriately weight terms often leads to ineffective retrieval. We show that re- writing the query to a version that com- prises a small subset of appropriate terms from the original query greatly improves effectiveness. Targeting a demonstrated potential improvement of almost 50% on some difficult TREC queries and their as- sociated collections, we develop a suite of automatic techniques to re-write queries and study their characteristics. We show that the shortcomings of automatic meth- ods can be ameliorated by some simple user interaction, and report results that are on average 25% better than the baseline. 1 Introduction Query expansion has long been a focus of infor- mation retrieval research. Given an arbitrary short query, the goal was to find and include additional related and suitably-weighted terms to the original query to produce a more effective version. In this pa- per we focus on a complementary problem – query re-writing. Given a long query we explore whether there is utility in modifying it to a more concise ver- sion such that the original information need is still expressed. The Y!Q beta1 search engine allows users to se- lect large portions of text from documents and issue them as queries. The search engine is designed to encourage users to submit long queries such as this example from the web site “I need to know the gas mileage for my Audi A8 2004 model”. The moti- vation for encouraging this type of querying is that longer queries would provide more information in the form of context (Kraft et al., 2006), and this ad- ditional information could be leveraged to provide a better search experience. However, handling such long queries is a challenge. The use of all the terms from the user’s input can rapidly narrow down the set of matching documents, especially if a boolean retrieval model is adopted. While one would ex- pect the underlying retrieval model to appropriately assign weights to different terms in the query and return only relevant content, it is widely acknowl- edged that models fail due to a variety of reasons (Harman and Buckley, 2004), and are not suited to tackle every possible query. Recently, there has been great interest in personal- ized search (Teevan et al., 2005), where the query is modified based on a user’s profile. The profile usu- ally consists of documents previously viewed, web sites recently visited, e-mail correspondence and so on. Common procedures for using this large amount of information usually involve creating huge query vectors with some sort of term-weighting mecha- nism to favor …

Content truncated to 500 words