Can Short Queries Be Even Shorter?

It is well known that query formulation could affect retrieval performance. Empirical observations suggested that a query may contain extraneous terms that could harm the retrieval effectiveness. This is true for both verbose and title queries. Given a query, it is possible that using its subqueries can generate more satisfying search results than using the original query. Although previous studies proposed method to reduce verbose queries, it remains unclear how we could reduce title queries given the short length of the title queries. In this paper, we focus on identifying the best performed subqueries for a given query. In particular, we formulate this problem as a ranking problem, where the goal is to rank subqueries of the query based on its predicted retrieval performance. To tackle this problem, we propose a set of novel post-retrieval features that can better capture relationships among query terms, and apply a learning to-rank algorithm based on these features. Empirical results over TREC collections show that these new features are indeed useful in identifying the best subqueries.