Self-Supervised Learning Based Target-Aware Session Recommendation Algorithm

TL;DR: Self-Supervised Learning-Based Target-Aware Session Recommendation Algorithm (SSL-TA) alleviates data sparsity and captures user interests in the target session by aggregating preferences from multiple levels of user implicit relationships.

Abstract: Session-based recommender systems aim to predict user behavior in the next moment based on anonymous sessions. Existing approaches suffer from the limitations of sparse data while ignoring the rich diversity of information embedded within items, resulting in the inability to capture the complete user interest in the target session. In order to better model user interests, we propose to better model user interests, we proposes a self-supervised learning-based target-aware session recommendation algorithm (SSL-TA). On the one hand, the idea of self-supervised learning is introduced to alleviate data sparsity; on the other hand, aggregation operations are performed from multiple levels of user preference implicit relationships in the target session to achieve finer-grained user interest inference. The effectiveness of the proposed model is verified by outperforming other benchmark models in experiments on two public datasets.