Query Recommendation with TF-IQF Model and Popularity Factor

Query recommendation is a technique that provides better queries to help users to get the needed documents when the original query submitted by the user may be insufficient or imprecise to retrieve those. In this paper a novel method for query recommendation is proposed. It is different from traditional methods in two aspects: (1) it breaks URLs into independent tokens and uses a TF-IQF model to present the queries, and calculates the query similarity based on that model in further steps, while traditional query log related methods take the clicked URLs recorded in query log as whole; and (2) it introduces a query popularity factor. The popularity factor adds weight to the queries that receive more user clicks, with the assumption that the quality of these popular queries is proven by previous users. In our experiments based on real commercial search engine query logs, our method out performs others, which demonstrates the effectiveness of the proposed TF-IQF model and popularity factor.