Improved Query Suggestion by Query Search

At the Web Intelligence conference in 2009, Jiang, Zilles, and Holte introduced a novel approach to query suggestion based on query search (QSQS), as well as a system-centered evaluation method. For each potentially relevant document, QSQS creates a complex query—called a lexical alias for the document— that ranks the document in its top 20. A technique called Query Search then builds query suggestions by simplifying the lexical alias. The present paper improves the state of the art by proposing two new query suggestion systems, IQSQS and GQSQS. Both replace the generation of lexical aliases by a simpler and more effective term selection process. They differ in their control structure: IQSQS builds query suggestions separately for each potentially relevant document, GQSQS builds them for a set of documents at once. Both our new systems substantially outperform QSQS in the measures introduced by Jiang et al. to evaluate QSQS; we achieve improvements of up to 30 percent in these measures for short user queries and up to 100 percent for long user queries. We show empirically that query expansion, which forces the user’s query to be included in each suggested query, is significantly superior to allowing the system the freedom to include or exclude terms from the user’s query at its discretion.