Semantic-Aware Diffusion Model for Sequential Recommendation

Yaoqi Chen, Jianjin Zhang, Qi Chen, Weihao Han, Zhengxin Zeng, Mingzheng Li, Xue Wu, Yujing Wang, Hao Sun, Xu Tan, Weiming Zhang, Jiang Bian, Weiwei Deng, Feng Sun, Qi Zhang

Keywords: Diffusion Model, Sequential Recommendation

TL;DR: We incorporate item semantics into the diffusion model through a Semantic Fusion Layer to enhance its performance for sequential recommendation.

Abstract: Sequential recommendation aims to predict the next click for a particular user based on their historical interacted item sequences. Recently, diffusion-based methods have achieved the state-of-the-art performance in sequential recommendation. However, they fail to effectively utilize the rich semantic information embedded in items during the diffusion process to accurately guide the generation, leading to sub-optimal results. To address this limitation, we designed SDRec, a **S**emantic-aware **D**iffusion model for sequential **Rec**ommendation. Our model introduces a novel architecture, the Semantic Fusion Layer, which leverages the embedding table from the encoder to incorporate item semantics into the diffusion process through an attention mechanism. Together with the well-designed contrastive and generative losses, SDRec effectively utilizes the item semantics in diffusion model, unleashing the potential of sequential recommendation. Our experiments show that SDRec has over 10% relative gain with superior efficiency compared with existing methods.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9990