Short-Term Recommendation With Recurrent Neural Networks

Collaborative filtering is a popular recommender algorithm that leverages its predictions and recommendations on the ratings or behaviors of other users. However, it needs to make use of all the users' ratings or behaviors rather than the users' recent ratings or behaviors to predict items. Therefore, it might ignore the consumers' habits changing with time. In this paper, we build a recurrent neural network to address the problem concerning on a time sequence and use gated recurrent units in recurrent neural network. The network treats a user's recent ratings or behaviors as a sequence, and each hidden layer models a user's rating or behavior which is in order. Furthermore, we integrate the gated recurrent unit with back propagation neural network to increase the prediction accuracy. Finally, our methods get higher precision accuracy in experiments.