A Soft Actor-Critic Algorithm for Sequential Recommendation

Recently, it has become common knowledge that using reinforcement learning for a sequential recommendation, which predicts a user's next action, can improve recommendation performance. This is because reinforcement learning can be used to efficiently learn behavioral changes, which can help you better understand user behavior patterns. Previous research has attempted to incorporate dynamic user characteristics through Actor-Critic algorithms, but these methods are limited in their ability to adequately learn user behavior because they learn without distinguishing between past and present behavior. Therefore, in this study, we propose a framework that incorporates the SAC algorithm, a reinforcement learning technique, to identify correlations between users and items in a dynamic environment where the recommender system continuously receives the next time series of data. Our framework outperformed from the viewpoint of the accuracy in the recommender system compared with the existing methods, and we could confirm that the SAC algorithm has the potential to improve the quality of the sequential recommendations in capturing the temporal dynamics of user interactions.