| 摘 要: 为缓解短序列用户数据稀疏导致的推荐不均衡问题,提出一种面向稀疏用户的公平推荐模型IAAW-CL,以融合信息增强与自适应加权对比学习策略。通过自注意力机制建模序列,并增强短序列用户的表示能力,生成多样化正样本视图,提升用户表征鲁棒性,并动态平衡推荐与对比学习目标。实验结果表明:相较于CL4SRec,IAAW-CL在Beauty数据集上,HR@5提升15.0%,NDCG@5提升10.1%;在Sports数据集上,HR@5提升16.9%,NDCG@5提升19.9%。模型有效缩小长短序列用户间的推荐差距,提升了推荐公平性。 |
| 关键词: 公平序列推荐 数据稀疏 信息增强 自注意力网络 对比学习 |
|
中图分类号: TP311
文献标识码: A
|
| 基金项目: 国家社会科学基金青年项目(25CGL002) |
|
| Fair Sequential Recommendation for Sparse Users:Contrastive Learning with Information Augmentation and Adaptive Weighting |
|
YE Shuchang1 , YIN Pei1,2 , JIANG Zhijia1
|
(1.Business School, University of Shanghai for Science and Technology, Shanghai 200093, China;
2.School of Intelligent Emergency Management, University of Shanghai for Science and Technology, Shanghai 200093, China)
yeashuc@ 163.com; pyin@ usst.edu.cn; 3108131774@ qq.com
|
| Abstract: To alleviate the recommendation imbalance caused by data sparsity in shor-t sequence users, a fairness- oriented recommendation model for sparse users is proposed, namely IAAW-CL (Information Augmentation and Adaptive Weighting for Contrastive Learning), integrating information augmentation and adaptive weighted contrastive learning strategies. The model leverages sel-f attention mechanisms to model behavior sequences and enhances the representation capability of shor-t sequence users by generating diverse positive sample views, improving the robustness of user representations, and dynamically balancing recommendation and contrastive learning objectives. Experimental results show that compared to Contrastive Learning for Sequential Recommendation (CL4SRec), IAAW-CL improves HR @ 5 by 15.0% and NDCG@ 5 by 10.1% on the Beauty dataset, and improves HR@ 5 by 16.9% and NDCG@ 5 by 19.9% on the Sports dataset. The model effectively reduces the recommendation gap between long and shor-t sequence users, enhancing recommendation fairness. |
| Keywords: fair sequential recommendation data sparsity information augmentation sel-f attention network contrastive learning |
