Contrastive Learning-Based Sequential Recommendation Model

Contrastive learning has demonstrated remarkable performance in sequential recommendation tasks. The core principle of it is to maximize consistency between the original and augmented data while increasing the distance between the original data and other instances, thereby facilitating the learning of more discriminative features. Nonetheless, such an approach has limitations and may inadvertently augment the distance between samples that belong to the same category in the feature space. To tackle the above problems, we propose a novel contrastive learning paradigm for sequential recommendation, termed CLSRec (Contrastive Learning-Based Sequential Recommendation Model). Our framework integrates cross-domain and intra-domain features through a meticulously designed contrastive loss function and introduces gray-scale positive sampling, aiming to address the inadequacy of focus on similar items in sequential recommendation. The proposed contrastive learning framework effectively captures intra-sequence item transition patterns and inter-sequence dependencies among items. Empirical evaluations on real-world datasets show that our model significantly outperforms advanced baseline models, validating its effectiveness in sequential recommendation scenarios.