A Knowledge Graph Recommendation Model Based on Reinforcement Learning and Comparative Learning

Recommendation system has become the core technology of recommendation platforms, with the continuous growth of user demand, how to improve the personalization and explanation quality of recommendation system has emerged as a significant area of research. This paper proposes a dynamic recommendation optimization method based on reinforcement learning and comparative learning which aims to solve the problems of insufficient flexibility of static weighting strategies and low quality of recommendation explanation in traditional recommendation systems. The method combines knowledge graph, optimizes the embedded representations of users and items through comparative learning, and introduces the Q-Learning algorithm to dynamically select positive-negative sample pairs and knowledge graph paths in order to improve recommendation accuracy and explanation transparency. The experimental findings demonstrate that this method considerably surpasses static optimization techniques and currently available knowledge graph-based recommendation models. After 10 rounds of training, the Accuracy (AUC) and F1 scores of the test set are both improved by about 3% over the existing models, and the Recall@K value is also higher than other models. In this paper, an adaptive recommendation framework integrating reinforcement learning and comparative learning is constructed, which provides theoretical support and practical foundation for more efficient personalized recommendations and diversified explanations.