Intent Contrastive Learning with Cross Subsequences for Sequential Recommendation
Published in WSDM 2024 (Oral), 2024
Recommended citation: Qin, Xiuyuan, Huanhuan Yuan, Pengpeng Zhao, Guanfeng Liu, Fuzhen Zhuang, and Victor S. Sheng. "Intent Contrastive Learning with Cross Subsequences for Sequential Recommendation." WSDM 2024. https://arxiv.org/abs/2310.14318
Venue: The 17th ACM International Conference on Web Search and Data Mining (WSDM 2024), Oral presentation.
Resources: Paper (arXiv) · Code · Video · Note
Abstract
User purchase behaviors are mainly influenced by their intentions. Modeling a user’s latent intention can significantly improve recommendation performance. Previous works model intentions via predefined auxiliary labels or stochastic data augmentation, which can be sparse, unavailable, or noisy. ICSRec segments sequential behaviors into multiple subsequences with a dynamic sliding operation and learns intention representations with coarse- and fine-grain intent contrastive learning. Experiments on four real-world datasets demonstrate superior performance over strong baselines.
Citation
Qin, Xiuyuan, Huanhuan Yuan, Pengpeng Zhao, Guanfeng Liu, Fuzhen Zhuang, and Victor S. Sheng. “Intent Contrastive Learning with Cross Subsequences for Sequential Recommendation.” WSDM 2024. arXiv:2310.14318.