Proximal Policy Optimization for Explainable Recommended Systems

Abstract

In this paper, in order to intuitively improve the interpretability of the recommendation, we make full use of the knowledge graph and provide an ‘explicit’ recommendation path for the recommended result. Specifically, we embeds the entities and relations in the knowledge graph according to consumers’ behavior. Then a knowledge reasoning method based on reinforcement learning algorithm is proposed, which aims to find a reasonable path after two entities in the knowledge graph are given. The main contributions of this paper are as follows: The current State-of-the-art method in reinforcement learning, PPO is used for improvement. The soft-reward function after adjustment behaves better for the current problem while training. And a Q-value based path exploration method is innovated for testing. With extensive experiments on several real-world datasets provided by Amazon, our method behaves better for the current problem compared with the baseline.