Neural Graph Collaborative Filtering

Neural Graph Collaborative Filtering (NGCF) targets a limitation of modern recommender systems: methods ranging from early matrix factorization to recent deep learning approaches typically build user and item embeddings by mapping from pre-existing features such as ID and attributes, which fails to encode the collaborative signal latent in user-item interactions. To address this, NGCF integrates the user-item interactions, specifically the bipartite graph structure, directly into the embedding process by propagating embeddings over the graph, enabling expressive modeling of high-order connectivity and explicitly injecting the collaborative signal into the embeddings.

Extensive experiments on three public benchmarks show that NGCF achieves significant improvements over several state-of-the-art models, including HOP-Rec and Collaborative Memory Network. Further analysis verifies the importance of embedding propagation for learning better user and item representations, justifying the rationality and effectiveness of explicitly modeling the user-item graph. The code was released publicly.