Contrastive Learning of Sequential Recommendation with Graph Attention Mechanisms

Sequential recommendation models the dynamic interest changes of users based on their interaction information with items, in order to predict the next interaction item for users. Existing sequential recommendation algorithms are susceptible to data sparsity issues. Contrastive learning sequential recommendation effectively alleviates data sparsity issues through construct samples and data enhancement methods, but it faces data noise problems. In response to the problem of introducing data noise in contrastive learning sequence recommendation, a graph attention mechanism-based contrastive learning sequential recommendation model (GACLRec) is proposed. This model uses the graph attention mechanism to dynamically evaluate and weight the importance of different neighboring nodes, effectively suppressing the interference of noise information and achieving interpretable aggregation of information from neighboring nodes. On this basis, in order to obtain richer node representations, the model further improved the multi-head graph attention mechanism to learn the diverse representations of nodes in multiple latent spaces. By integrating these representations from different spaces, a comprehensive node feature vector is finally obtained, enabling the model to comprehensively learn the actual preferences and behavior patterns of users. Compared with existing excellent algorithms on the real datasets, the beauty and yelp. The experiment proves that the model proposed in this paper has a certain improvement compared to the baseline method on the evaluation metrics HR@N and NDCG@N.