Neighbor Interaction Aware Graph Neural Network for Collaborative Filtering

Recently, recommendation algorithms based on Graph Neural Networks (GNNs) have demonstrated robust capabilities in learning representations of users and items. However, these methods often rely on aggregation functions like sum/mean/max pooling operations to form user and item representations, overlooking the importance of interaction information between neighboring nodes for node representation. To address these limitations, this paper introduces the Neighbor Interactions Aware Graph Neural Network (NIA-GNN) method for collaborative filtering. This approach explicitly models both the aggregation and interaction information of neighboring nodes, capturing relationship information between adjacent nodes to enhance user and item representations, improve recommendation effectiveness, and boost the model's predictive power. We perform extensive experiments on two real datasets to confirm that NIA-GNN significantly enhances recommendation performance. Further analysis underscores the significance of neighbor interaction information in refining user and item representations, validating the rationale and efficacy of the NIA-GNN method.