Mixing-RNN: A Recommendation Algorithm Based on Recurrent Neural Network

Enhan Liu, Yan Chu, Lan Luan, Guang Li, Zhengkui Wang

Abstract: Collaborative filtering algorithms have been used by recommender systems for item (e.g., movie) recommendation. However, traditional collaborative filtering algorithms face challenges to provide accurate recommendation when users’ interest and context suddenly changed. In this paper, we present a new Recurrent Neural Network-based model, namely Mixing-RNN that is able to capture time and context changes for item recommendation. In particular, Mixing-RNN integrates the insight from Rating-RNN and Category-RNN which are developed to predict users’ interest based on rating and category respectively. Different from the traditional RNN, we integrate the forget gate and input gate in the model, where the forget gate decides what information to remain or discard and the input gate inputs rating information to the model. Our experiment evaluation on MovieLens indicates that Mixing-RNN outperforms the state-of-art methods.