Path recommendation based on attention mechanism and multiple rewards

Traditional recommendation methods overlook diversity and personalized requirements in path selection. With the ongoing expansion of knowledge graph data and the diversification of application scenarios, there is a rising demand for customized preferences and diverse options in path recommendation. Therefore, how can attention mechanisms and multiple rewards be effectively integrated to better align with the users' needs? This paper proposes a path recommendation method based on an attention mechanism and multiple rewards (PR-AM&MR) to handle these issues. This method utilizes the attention mechanism to measure the weights of neighboring nodes to exploit the implicit relationships among different items deeply. Furthermore, it dynamically uses reinforcement learning to adjust the recommended path, updating its optimization strategy based on diverse reward outcomes and achieving real-time feedback to optimize the recommendation path. Experiments are conducted on the JD e-commerce and KKBOX datasets to evaluate the proposed method. Significant advantages are obtained compared with robust baseline methods.