A generative framework to query recommendation and evaluation

Abstract

In practice, query recommenders in Web search typically recommend queries directly from a query log or iteratively refine a user's current context to make recommendations. These approaches either limit themselves to queries in the log or do not take necessary exploratory leaps in their recommendations. Moreover, they do not directly incorporate the encompassing, driving information needs and tasks. The author first shows that user queries may not necessarily be the best to use for recommendations and moreover proposes a framework for generating novel queries for query recommendation, using an approximation of need. Copyright is held by the owner/author(s).