TAGCL: A Time-Aware Recommendation Method via Graph Contrastive Learning

In the era of information explosion, recommender systems serve as critical tools for filtering valuable content. While emerging graph contrastive learning (GCL) approaches introduced into recommendation enhance anti-noise robustness, temporal constraints are often neglected in user behavior sequences, leading to interest inference bias. Concurrently, the existing temporal recommendation models typically concentrate solely on temporal factors in single tasks but lack of the capacity to extract more profound temporal information. To address these limitations, this paper proposes a Time-Aware recommendation method via Graph Contrastive Learning (TAGCL) with two core contributions: (1) Introducing an innovative Time-aware EdgeDropping strategy (TED) for preserving temporal causality through interval-based edge pruning; (2) Constructing a multitask learning architecture jointly optimizing the GCL-based selfsupervised task and the GNN-based recommendation objective task to deepen temporal features of interaction behaviour. Conducting experiments and verifying through commonly used datasets demonstrate TAGCL's superiority over baselines, significantly improving noise robustness and temporal constraint modeling, which is specifically manifested in the recommendation performance up to 4.20% NDCG@5gain.