Neural Graph Collaborative Filtering (NGCF) is a recommendation framework that integrates user-item interactions into the embedding process to capture collaborative signals. The core idea is to propagate embeddings on the user-item graph to model high-order connectivity, thereby explicitly injecting collaborative signals into the embedding process. This approach improves the quality of user and item representations, leading to better recommendation performance. The framework is evaluated on three public benchmarks, demonstrating significant improvements over state-of-the-art models like HOPRec and Collaborative Memory Network. The results show that embedding propagation is crucial for learning effective representations, and NGCF outperforms other methods in terms of recall and NDCG metrics. The method is efficient and scalable, with a time complexity that is manageable for large-scale datasets. The study also highlights the importance of high-order connectivity in capturing collaborative signals and the effectiveness of embedding propagation in enhancing recommendation performance. The results indicate that NGCF achieves state-of-the-art performance on various datasets, demonstrating its effectiveness in capturing collaborative signals and improving recommendation accuracy.Neural Graph Collaborative Filtering (NGCF) is a recommendation framework that integrates user-item interactions into the embedding process to capture collaborative signals. The core idea is to propagate embeddings on the user-item graph to model high-order connectivity, thereby explicitly injecting collaborative signals into the embedding process. This approach improves the quality of user and item representations, leading to better recommendation performance. The framework is evaluated on three public benchmarks, demonstrating significant improvements over state-of-the-art models like HOPRec and Collaborative Memory Network. The results show that embedding propagation is crucial for learning effective representations, and NGCF outperforms other methods in terms of recall and NDCG metrics. The method is efficient and scalable, with a time complexity that is manageable for large-scale datasets. The study also highlights the importance of high-order connectivity in capturing collaborative signals and the effectiveness of embedding propagation in enhancing recommendation performance. The results indicate that NGCF achieves state-of-the-art performance on various datasets, demonstrating its effectiveness in capturing collaborative signals and improving recommendation accuracy.