Multi-Aspect enhanced Graph Neural Networks for recommendation.

Graph neural networks (GNNs) have achieved remarkable performance in personalized recommendation, for their powerful data representation capabilities. However, these methods still face several challenging problems: (1) the majority of user-item interaction graphs only utilize the interaction information, which cannot reflect the users' specific preferences for different aspects, making it difficult to capture user preferences in a fine-grained manner. (2) there is no effective way to integrate multi-aspect preferences into a unified model to capture the comprehensive user interests. To address these challenges, we propose a Multi-Aspect enhanced Graph Neural Networks (MA-GNNs) model for item recommendation. Specifically, we learn the aspect-based sentiments from reviews and use them to construct multiple aspect-aware user-item graphs, thus giving the edge practical meaning. And aspect semantic features are introduced into the information aggregation process to adjust users' preferences for different items. Furthermore, we design a routing-based fusion mechanism, which adaptively allocates weights to different aspects to realize the dynamic fusion of aspect preferences. We conduct experiments on four publicly available datasets, and the experimental results show that the proposed MA-GNNs model outperforms state-of-the-art methods. Further analysis proves that fine-grained interest modeling can improve the interpretability of recommendations.