Query ranking model for search engine query recommendation

In this paper, we propose a query ranking model to select and order queries for search engine query recommendations. In contrast to existing similarity-based query recommendation methods (Agglomerative clustering of a search engine query log, 2000; The query-flow graph: model and applications, 2008], this model is based on utility, and ranks a query based on the joint probability of events whereby a query is selected by the user, the search results of the query are selected by the user, and the chosen search results satisfy the user's information needs. We thus define three utilities in our model: a query-level utility representing the attractiveness of a query to the user, a perceived utility measuring the user's actions given the search results, and a posterior utility measuring the user's satisfaction with the chosen search results. We propose methods to compute these three utilities from query log data. In experiments involving real query log data, our proposed query ranking model outperformed seven other baseline methods in generating useful recommendations.