Using Multiple Query Expansion Algorithms to Predict Query Performance

Query Performance Prediction (QPP) may be defined as the problem of predicting the effectiveness of a search system for a given query and a collection of documents without any relevance judgments. QPP is a useful problem to study: if the performance of a search system for a given query can be estimated in advance (or during retrieval), it may be possible to tailor the retrieval strategy so that the overall effectiveness of the system is improved. This paper presents a novel method for QPP that is based on Query Expansion (QE) algorithms. The proposed method uses the overlap between two expanded versions of the same query (expanded using two different classes of QE algorithms) as a query performance predictor. The method itself does well on certain test collections, but is noticeably inferior to a state of the art method like NQC on other datasets. To leverage the complementary behaviour of our method and NQC, we combine the two methods with excellent results across a number of standard test collections.