Graph convolutional neural network recommendation system with feature enhancement based on collaborative filtering

• Yang Xu,
• Yueyi Zhang,
• Hanting Zhou,
• Jing Hu,

Vol. 19, No. 11, November 30, 2025
10.3837/tiis.2025.11.009, Download Paper (Free):

• Recommendation System
• Graph Convolutional Neural Network
• heterogeneous graph
• Collaborative Filtering
• Edge node embedding

Abstract

In recent years, recommendation systems (RSs) have become increasingly crucial in various domains such as e-commerce, streaming platforms, and social media. Graph neural networks (GNNs) have emerged as predominant approaches in this field. However, existing GNN-based recommendation systems (RSs) often face critical limitations. These methods typically depend on basic node features, such as position indices or raw rating matrices, which lack sufficient representational power. Moreover, there exist data sparsity and cold-start problems that are prevalent in real-world recommendation scenarios. To overcome these limitations, this paper proposes a novel collaborative filtering enhanced graph convolutional neural network recommendation system (GCN-CF). GCN-CF enhances node representation through matrix factorization and effectively solves data sparsity and cold start problems through graph convolutional networks. GCN-CF includes three main phases: MF-based feature extraction from user-item interactions, GCN-based training on heterogeneous graphs (user-item, user-user, and item-item graphs) to obtain enhanced edge node embeddings, and rating prediction through a linear regressor. Extensive experiments conducted on three benchmark datasets (MovieLens 100K, MovieLens 1M, and Beauty) demonstrate that GCN-CF achieves significant performance improvements. The results confirm that our approach, which combines matrix factorization (MF) based feature extraction with bilateral domain information fusion, effectively enhances model performance. Furthermore, the use of MF-initialized embedding features reduces the model's computational runtime. Code link: https://zenodo.org/records/15240020.

Statistics

Cite this article