Recommendation Method Based on Multi-task Learning and Graph Collaborative Filtering

The rapid growth of Internet technology is continuously increasing information volume, leading to the problem of information overload. When faced with a massive amount of information, the focus is on how to quickly and accurately find the desired information, which has led to the emergence of recommendation systems. However, traditional collaborative filtering recommendations are often affected by issues such as cold start and data sparsity, while existing graph-based collaborative filtering methods also suffer from poor representation learning quality and data sparsity. So we propose a recommendation method based on Multi-Task Learning and Graph Collaborative Filtering (MTLGCF). This method uses graph collaborative filtering recommendation as the main task, incorporates two contrastive learning tasks as auxiliary tasks, and introduces a data augmentation strategy, enabling the model to perform multi-task learning. By evaluating on three public datasets such as ml-latest-small and comparisons with nine recent recommendation methods, the proposed MTLGCF model shows significant improvements in Recall and NDCG metrics.