Beyond Query Logs: Recommendation and Evaluation

Abstract

Query recommendation in Web search is typically manifested in algorithms that 1) recommend previously issued queries from a query log or 2) make incremental changes to queries in a user's current session. While such approaches have been effective in improving retrieval, they either are limited to suggesting queries in a query log or fail to make appropriate leaps that are necessary for query recommendation. More crucially, these approaches only recommend queries that are a coarse approximation of the information a user needs to complete their goal. They do not directly attempt to model the need and generate recommendations from it. This work will propose a framework for generating novel yet focused queries for query recommendation.