The Role of the User's Browsing and Query History for Improving MPC-generated Query Suggestions

Abstract

In this paper we use the user’s recent web browsing history in order to provide better query suggestions in an information retrieval system. We have built a Chrome browser plugin that collects each web page visited by a user and submits it to our query suggestion server for indexing, thus building a personal history profile for each user. We then analyze if future queries submitted by a user to the search engine can be predicted from web pages visited by that user in the past (i.e. his recent browsing history) or from queries submitted by that user in the past (i.e. his recent query history). The contribution of this paper is a method of using this personal history profile for reordering the query suggestions offered by Google when the user searches information on Google, moving query suggestions more relevant to the user’s information need to the front positions in the Google provided query suggestions list. We have collected browsing history log data for over 4 months from several users who installed our Chrome plugin on their local computers and then we performed an offline evaluation test that shows that our personalized query suggestion system improves the MRR (i.e. Mean Reciprocal Rank) score of Google query suggestions by approximately 0.04 (i.e. improves Google’s MRR score by 12 percents).

Keywords