Predicting query performance in domain-specific corpora

The performance of a document recommender system is dependent on the quality and characteristics of the query used by the recommender to retrieve relevant documents. Automatically predicting the performance of a query can help identify ineffective queries and can help improve performance by selectively applying query expansion techniques. In this paper, we study Information-entropy-based measures for predicting performance of a query in the context of domain-specific corpora. We propose a new sampling mechanism that can more accurately estimate query models in domain-specific corpora and hence deliver better predictions. We evaluate the validity our technique by analyzing its performance in five different domain-specific corpora. © 2007 IEEE.