Collaborative Graph Neural Networks with Contrastive Learning for Sequential Recommendation

Sequential recommendations aim to exploit user purchase records to predict the next items they will buy. Although the problem has been extensively investigated, most existing methods model user interests based only on their own historical sequences but ignore the collaboration with others. Additionally, user behaviors are often implicit and contain noise that cannot fully reflect their preferences. Furthermore, since user preferences can evolve over time, capturing their interests from past behaviors becomes even more difficult. To address the above issues, we propose a novel method called Collaborative Graph Neural Networks with Contrastive learning (C2GNN) for sequential recommendations. Specifically, our approach leverages a method that incorporates user activity, item popularity, and interaction time to construct dynamic subgraphs from multiple user sequences, which are fed into a graph neural network (GNN) for feature aggregation. To further enhance the model's discriminative ability, we introduce a contrastive learning framework that learns user and item representations by comparing different views of the GNN output at different layers. Extensive experiments on three real-world datasets show that our proposed C2GNN method outperforms state-of-the-art methods for sequential recommendations. Our code and data are published at https://github.com/tabo0/CCGNN.