Improving Graph Collaborative Filtering with Neighborhood-enriched Contrastive Learning

Recently, graph collaborative filtering methods have been proposed as an recommendation approach, which can capture users' preference over by modeling the user-item interaction graphs. In order to reduce the of data sparsity, contrastive learning is adopted in graph filtering for enhancing the performance. However, these methods construct the contrastive pairs by random sampling, which neglect the relations among users (or items) and fail to fully exploit the of contrastive learning for recommendation. To tackle the above, we propose a novel contrastive learning approach, named-enriched Contrastive Learning, named NCL, which explicitly the potential neighbors into contrastive pairs. Specifically, we the neighbors of a user (or an item) from graph structure and space respectively. For the structural neighbors on the interaction, we develop a novel structure-contrastive objective that regards users (or items) and their structural neighbors as positive contrastive pairs. In, the representations of users (or items) and neighbors to the outputs of different GNN layers. Furthermore, to excavate the neighbor relation in semantic space, we assume that users with representations are within the semantic neighborhood, and incorporate semantic neighbors into the prototype-contrastive objective. The proposed can be optimized with EM algorithm and generalized to apply to graph filtering methods. Extensive experiments on five public datasets the effectiveness of the proposed NCL, notably with 26% and 17% gain over a competitive graph collaborative filtering base model on Yelp and Amazon-book datasets respectively. Our code is available at: : //github. com/RUCAIBox/NCL.