A New Deep Neural Network Based Learning to Rank Method for Information Retrieval
Mingsheng Fu, Hong Qu, Fan Li, Yan‐Jun Liu · 2018
Applying deep neural network to learning to rank is an attractive emerging issue in information retrieval (IR). However, most of existing deep neural network based methods cannot even beat traditional methods regarding ranking precision, since the adopted raw text of query and documents contain a significant amount of irrelevance words. To overcome this problem, we propose a new deep neural network based method directly using extracted IR features, which is more robust and correlated to the final ranking task. Besides, our method takes both local and global perspectives on documents and query into account to produce an accurate ranking. Consequently, two separated neural networks are involved in considering the local and global views explicitly, and then the ranking decision is provided by the interaction between the extracted features of these networks. Experiments on two popular benchmarks MQ2007 and MQ2008 are carried out to verify the performance of the proposed method, and the experimental results show our approach significantly outperforms all of the state-of-the-art methods.