Recurrent Coevolutionary Feature Embedding Processes for Recommendation

Hanjun Dai*, Yichen Wang*, Rakshit Trivedi, Le Song

Abstract: Recommender systems often use latent features to explain the behaviors of users and capture the properties of items. As users interact with different items over time, user and item features can influence each other, evolve and co-evolve over time. To accurately capture the fine grained nonlinear coevolution of these features, we propose a recurrent coevolutionary feature embedding process model, which combines recurrent neural network (RNN) with a multi-dimensional point process model. The RNN learns a nonlinear representation of user and item embeddings which take into account mutual influence between user and item features, and the feature evolution over time. We also develop an efficient stochastic gradient algorithm for learning parameters. Experiments on diverse real-world datasets demonstrate significant improvements in user behavior prediction compared to state-of-the-arts.

TL;DR: Our work combines recurrent neural network with point process models for recommendation, which captures the co-evolution nature of users' and items' latent features.

Conflicts: gatech.edu

Keywords: Deep learning, Applications