Deep Generative Recommendation with Maximizing Reciprocal Rank

Abstract

Variational autoencoders (VAEs) have proven to be successful in the field of recommender systems. The advantage of this non-linear probabilistic generative model is that it can break through the limited modeling capabilities of linear models which dominate collaborative filtering research to a large extend. In this paper, we propose a deep generative recommendation model by enforcing a list-wise ranking strategy to VAE with the aid of multinomial likelihood. This model has ability to simultaneously generate the point-wise implicit feedback data and create the list-wise ranking list for each user. To seamlessly combine ranking loss with VAE loss, the Reciprocal Rank (RR) is adopted here and approximated with a smoothed function. A series of experiments on two real-world datasets (MovieLens-100k and XuetangX) have been conducted. We show that maximizing the ranking loss will cause as many relevant items appearing at the top of the predicted recommendation list as possible. The experimental results demonstrated that the proposed method outperforms several state-of-the-art methods in ranking estimation task.

This work was supported in part by the National Natural Science Foundation of China under Grant 61822601, 61773050, and 61632004; the Beijing Natural Science Foundation under Grant Z180006; National Key Research and Development Program (2017YFC1703506); The Fundamental Research Funds for the Central Universities (2019JBZ110).