Enhanced Attention-based Self-Supervised Co-Training for Session-based Recommendation
Session-based recommendation struggles with data sparsity, and existing self-supervised methods often discard crucial information. We propose EASCOT, a dual-encoder framework leveraging item and session graphs with contrastive learning. By preserving full session data and integrating Transformer modeling, EASCOT significantly improves performance on benchmarks.
Authors
- Z. S. Zhang (ORCID: https://orcid.org/0009-0007-7858-3575)
- Weinan Jiang
- Hualin Zhan
Institutions
- Jiangxi Science and Technology Normal University (CN)
Publication Details
- Published
- 2025-04-11
- DOI
- https://doi.org/10.1109/ainit65432.2025.11035113
- Primary Topic
- Recommender Systems and Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00