Diffusion Augmentation for Sequential Recommendation

By Q. Liu et al

Published on Oct. 21, 2025

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Preliminary
2.1 Problem Definition and Notations
2.2 Diffusion Model
3. DiffuASR
3.1 Framework
3.2 SU-Net
3.3 Guide Procedure
4. Conclusion

Summary

Sequential recommendation (SRS) has become the technical foundation in many applications recently, aiming to recommend the next item based on the user's historical interactions. The problem of data sparsity and the long-tail user problem have hindered the development of sequential recommendation. To address these challenges, a Diffusion Augmentation for Sequential Recommendation (DiffuASR) framework is proposed, leveraging the diffusion model to augment data quality. The framework includes a forward process for embedding sequence, a reverse process for recovery, and a guide procedure for user preference assimilation. A Sequential U-Net (SU-Net) architecture is designed to capture the sequential information of the embedding sequence. Extensive experiments illustrate the effectiveness and generality of DiffuASR in enhancing sequential recommendation models.