Diffusion-based Long and Short Term Interest Sequence Recommendation

Xiaowen Wang, Thomas Tran
Proceedings of the The 39th Canadian Conference on Artificial Intelligence, PMLR 318:856-863, 2026.

Abstract

Sequential recommendation requires modeling both stable long-term preferences and dynamic short-term intents. However, most existing methods rely on static fusion strategies, which cannot adaptively balance these signals. To address this, we propose DiffLSRec, a diffusion-based framework that performs progressive fusion of long- and short-term representations. The long-term embedding is treated as a prior, while short-term intent provides guidance during multi-step denoising, enabling dynamic and fine-grained integration. We further enhance short-term modeling with token-level contextual information and regulate the fusion process using SNR-adaptive guidance. Experiments on three Amazon datasets show that DiffLSRec consistently outperforms representative baselines across multiple metrics.

Cite this Paper

BibTeX
@InProceedings{pmlr-v318-wang26a, title = {Diffusion-based Long and Short Term Interest Sequence Recommendation}, author = {Wang, Xiaowen and Tran, Thomas}, booktitle = {Proceedings of the The 39th Canadian Conference on Artificial Intelligence}, pages = {856--863}, year = {2026}, editor = {Bouzar-Benlabiod, Lydia and Leung, Carson}, volume = {318}, series = {Proceedings of Machine Learning Research}, month = {25--29 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v318/main/assets/wang26a/wang26a.pdf}, url = {https://proceedings.mlr.press/v318/wang26a.html}, abstract = {Sequential recommendation requires modeling both stable long-term preferences and dynamic short-term intents. However, most existing methods rely on static fusion strategies, which cannot adaptively balance these signals. To address this, we propose DiffLSRec, a diffusion-based framework that performs progressive fusion of long- and short-term representations. The long-term embedding is treated as a prior, while short-term intent provides guidance during multi-step denoising, enabling dynamic and fine-grained integration. We further enhance short-term modeling with token-level contextual information and regulate the fusion process using SNR-adaptive guidance. Experiments on three Amazon datasets show that DiffLSRec consistently outperforms representative baselines across multiple metrics.} }
Endnote
%0 Conference Paper %T Diffusion-based Long and Short Term Interest Sequence Recommendation %A Xiaowen Wang %A Thomas Tran %B Proceedings of the The 39th Canadian Conference on Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Lydia Bouzar-Benlabiod %E Carson Leung %F pmlr-v318-wang26a %I PMLR %P 856--863 %U https://proceedings.mlr.press/v318/wang26a.html %V 318 %X Sequential recommendation requires modeling both stable long-term preferences and dynamic short-term intents. However, most existing methods rely on static fusion strategies, which cannot adaptively balance these signals. To address this, we propose DiffLSRec, a diffusion-based framework that performs progressive fusion of long- and short-term representations. The long-term embedding is treated as a prior, while short-term intent provides guidance during multi-step denoising, enabling dynamic and fine-grained integration. We further enhance short-term modeling with token-level contextual information and regulate the fusion process using SNR-adaptive guidance. Experiments on three Amazon datasets show that DiffLSRec consistently outperforms representative baselines across multiple metrics.
APA
Wang, X. & Tran, T.. (2026). Diffusion-based Long and Short Term Interest Sequence Recommendation. Proceedings of the The 39th Canadian Conference on Artificial Intelligence, in Proceedings of Machine Learning Research 318:856-863 Available from https://proceedings.mlr.press/v318/wang26a.html.

Related Material