【Reprinted from SJTU News】

SIGIR 2017 will be held in Tokyo, Japan from August 7 to 11. It is a top academic conference in the field of information retrieval. “IRGAN: A Minimax Game for Unifying Generative and Discriminative Information Retrieval Models” has been accepted with three reviews of “strong accept”, which is the best evaluation among the 362 papers submitted to the conference.

Prof. Wang Jun from University College London is the first author of the paper, while Assistant Prof. Zhang Weinan from our department is the corresponding author who presided over all of the experiments. 

The paper provides a unified account of two schools of thinking in information retrieval modelling: the generative retrieval focusing on predicting relevant documents given a query, and the discriminative retrieval focusing on predicting relevancy given a query-document pair. It proposes a game theoretical minimax game to iteratively optimise both models. It shows that the unified framework takes advantage of both schools of thinking: (i) the generative model learns to fit the relevance distribution over documents via the signals from the discriminative model, and (ii) the discriminative model is able to exploit the unlabelled data selected by the generative model to achieve a better estimation for document ranking. The experimental results have demonstrated significant performance gains as much as 23.96% on Precision@5 and 15.50% on MAP over strong baselines in a variety of applications including web search, item recommendation, and question answering.