Query Auto-Completion for Rare Prefixes

Proceedings of the 24th ACM International Conference on Information and Knowledge Management (CIKM) |

Published by ACM - Association for Computing Machinery

Publication | Publication

Query auto-completion (QAC) systems typically suggest queries that have previously been observed in search logs. Given a partial user query, the system looks up this query prefix against a precomputed set of candidates, then orders them using ranking signals such as popularity. Such systems can only recommend queries for prefixes that have been previously seen by the search engine with adequate frequency. They fail to recommend if the prefix is sufficiently rare such that it has no matches in the precomputed candidate set.

We propose a design of a QAC system that can suggest completions for rare query prefixes. In particular, we describe a candidate generation approach using frequently observed query suffixes mined from historical search logs. We then describe a supervised model for ranking these synthetic suggestions alongside the traditional full-query candidates. We further explore ranking signals that are appropriate for both types of candidates based on n-gram statistics and a convolutional latent semantic model (CLSM). Within our supervised framework the new features demonstrate significant improvements in performance over the popularity-based baseline. The synthetic query suggestions complement the existing popularitybased approach, helping users formulate rare queries.