Building the multi-objective periodic recommendation system through integrating optimistic linear support and user clustering to multi-object reinforcement learning

Our study focuses on the diversity of user preferences and the dynamics of the user-product relationship, particularly in the context of periodic product usage. The principal objective of this research is to explore multi-objective optimization for a recommendation system tailored to periodic products. Our methodology employs a multi-objective reinforcement learning (MORL) algorithm. Additionally, we have proposed integrating the optimistic linear support algorithm into a MORL algorithm to collect good weight vectors. We also proposed using user clustering to ensure the model remembers user’s preferences in early episodes. The findings of this research demonstrate that our proposed multi-objective approach yields significantly higher effectiveness when contrasted with conventional single-objective methodologies.