Personalized neural language models for real-world query auto completion

  • 2018-04-17 19:11:14
  • Nicolas Fiorini, Zhiyong Lu
  • 7

Abstract

Query auto completion (QAC) systems are a standard part of search engines inindustry, helping users formulate their query. Such systems update theirsuggestions after the user types each character, predicting the user's intentusing various signals - one of the most common being popularity. Recently, deeplearning approaches have been proposed for the QAC task, to specificallyaddress the main limitation of previous popularity-based methods: the inabilityto predict unseen queries. In this work we improve previous methods based onneural language modeling, with the goal of building an end-to-end system. Weparticularly focus on using real-world data by integrating user information forpersonalized suggestions when possible. We also make use of time informationand study how to increase diversity in the suggestions while studying theimpact on scalability. Our empirical results demonstrate a marked improvementon two separate datasets over previous best methods in both accuracy andscalability, making a step towards neural query auto-completion in productionsearch engines.

 

Quick Read (beta)

loading the full paper ...