SRRL: Select Reliable Friends for Social Recommendation with Reinforcement Learning

摘要

Along with the boom of social media, researchers consider that incorporating social relationships into traditional recommender systems can help alleviate the problems of data sparsity and cold start. However, recent reports show that social recommendation can hardly get the expected effect. The unsatisfying result can be attributed to the fact that users may share different preferences with their friends because of the randomness in the process of building social networks. In consequence, direct use of social relationships may lead to the degradation of recommendation performance, which makes identifying reliable friends for each user critical. In this paper, we propose an end-to-end social recommendation framework based on reinforcement learning to identify reliable social relationships for users. Specifically, our model consists of two parts: (a) an agent which samples users' reliable social relationships and delivers them to the environment, (b) an environment which takes charge of generating recommendations with sampled social relations and returning rewards to the agent to optimize the sampling procedure. With the interactions between the agent and the environment, our model can dynamically identify reliable friends whose preferences are really similar to the current user. Experimental analysis on two real-world datasets demonstrates that our approach outperforms the state-of-the-art social recommendation algorithms.