Quantum Language Model-based Query Expansion

Abstract

The analogy between words, documents and queries and the Quantum Mechanics (QM) concepts gives rise to various quantum-inspired Information Retrieval (IR) models. As one of the most successful applications among them, Quantum Language Model (QLM) achieves superior performances compared to various classical models on ad-hoc retrieval tasks. However, the EM-based estimation strategy for QLM is limited in that it cannot efficiently converge to global optimum. As a result, subsequent QLM-based models are more or less restricted to a limited vocabulary. In order to ease this limitation, this study investigates a query expansion framework on the QLM basis. Essentially, the additional terms are selected from the constructed QLM of top-K returned documents in the initial ranking, and a re-ranking is conducted on the expanded query to generate the final ranks. Experiments on TREC 2013 and 2014 session track datasets demonstrate the effectiveness of our model.