Multi-view knowledge graph convolutional networks for recommendation

Recommender systems based on knowledge graphs (KGs) have attracted increasing attention recently, which alleviates the sparsity and cold-start issues by modeling user-item interactions with side information. However, most KG-based recommendation systems focus on shallow models due to the over-smoothing issue and neglect users' long-term preferences. Moreover, KG-based methods are susceptible to noisy interactions, which reduces the robustness of the recommendation system. In this work, we propose a recommendation model based on a multi-view knowledge graph convolutional network (MKGCN) to address these issues. Specifically, to mitigate the impact of noisy interactions in a KG, we construct multi-view knowledge graphs from raw data by random sampling to learn multiple representations. Moreover, we develop a multi-layered knowledge graph convolutional network by introducing the initial residual connection to alleviate the over-smoothing issue. This approach enables effective capture of high-order connectivity and exploration of users' potential long-term preferences. Furthermore, the graph self-attention mechanism is utilized to filter out inherent noise and refine the recommended results. Experiment results on real-world datasets demonstrate the effectiveness of MKGCN and its superiority over several state-of-the-art methods.