Scholarlink

<p>Next-basket recommendation (NBR) aims to recommend items to a user based on his/her historical basket sequence. Existing methods typically employ neural networks to model a user’s historical basket sequence. While achieving promising results, these methods overlook the data sparsity problem inherent in NBR. Data sparsity problem prevents models from obtaining sufficient training. To alleviate the data sparsity problem in NBR, we propose a multi-task learning framework named Contrastive Learning for Next-basket Recommendation (CL4NBR). CL4NBR introduces a contrastive learning task alongside the original next-basket prediction task, leveraging self-supervised signals from users’ historical basket sequences. This approach enables the generation of more effective user representations, leading to more accurate recommendation results. To obtain self-supervised signals from users’ basket sequences, three data augmentation methods for basket sequences are designed in this study. Furthermore, CL4NBR utilizes the repeated purchase information of users when predicting their next baskets. Experimental evaluations conducted on four public datasets demonstrate that CL4NBR outperforms state-of-the-art methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Contrastive learning for next-basket recommendation

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Shaoqiang Zhu,
  • Qiang Wang,
  • Kejia Zhang,
  • Haiwei Pan

摘要

Next-basket recommendation (NBR) aims to recommend items to a user based on his/her historical basket sequence. Existing methods typically employ neural networks to model a user’s historical basket sequence. While achieving promising results, these methods overlook the data sparsity problem inherent in NBR. Data sparsity problem prevents models from obtaining sufficient training. To alleviate the data sparsity problem in NBR, we propose a multi-task learning framework named Contrastive Learning for Next-basket Recommendation (CL4NBR). CL4NBR introduces a contrastive learning task alongside the original next-basket prediction task, leveraging self-supervised signals from users’ historical basket sequences. This approach enables the generation of more effective user representations, leading to more accurate recommendation results. To obtain self-supervised signals from users’ basket sequences, three data augmentation methods for basket sequences are designed in this study. Furthermore, CL4NBR utilizes the repeated purchase information of users when predicting their next baskets. Experimental evaluations conducted on four public datasets demonstrate that CL4NBR outperforms state-of-the-art methods.