Attention-based Deep Reinforcement Learning Model for Pair-Wise Interaction Recommendation
Chenyan Zhang, Huanmei Guan, Ni Li · 2019
Traditional recommendation algorithms generate a recommendation list for the user by rating prediction or ranking based on their historical record. However, most of these methods still faced with two problems: (1) only considering users' positive feedback; (2) seeking to maximize the immediate recommendation effect. To address the above issues, we propose a novel deep reinforcement learning based interactive recommendation model, named DDQN-AM. Specifically, we first use long short-term memory network to capture positive and negative interest signals from user's historical behavior as Deep Double Q-Network(DDQN)'s input states. Secondly, we also propose an attentional mechanism(AM) to learn the effects of contextual information for weight calculation. Extensive experiments on real data show that the proposed DDQN-AM model outperforms the state-of-the-art algorithms, and verify the importance of attention mechanism.