KSRG:Knowledge-Aware Sequential Recommendation with Graph Neural Networks

Abstract

In the actual recommendation scenario, user’s next action is affected by their long-term intrinsic preference and short-term temporal demands. Therefore, how to model user’s long-term and short-term preferences has become the key to improving the accuracy of the recommendation algorithm. However, the current mainstream Markov Chains and Recurrent Neural Networks can not well capture the dynamic interests of users. In view of this, we propose Knowledge-Aware Sequential Recommendation with Graph Neural Networks (KSRG), a novel sequential recommendation framework that takes both semantic information and temporal dynamics into consideration. Specifically, Graph Neural Network (GNN) is utilized to effectively capture the global graph structure information, which reflects user’s long-term interests. In order to capture user short-term preferences, we use relational intensity and Fourier-based temporal decay function to estimate the impact of historical interaction on the target item. Finally, the combined long-term and short-term preferences of users will help to generate a recommendation list. Experimental studies on real-world datasets demonstrate that our method outperforms many strong baselines.