Reproduce “Sequential Recommendation with Diffusion Models” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hanwen Du, Huanhuan Yuan, Zhen Huang, Pengpeng Zhao, Xiaofang Zhou

Soochow University and HKUST researchers developed DiffRec, a sequential recommendation framework utilizing diffusion models, to address limitations of prior generative methods like VAEs and GANs. DiffRec achieved state-of-the-art performance, outperforming baselines by up to 25.94% in NDCG@5 on ML-1M and 16.27% in HR@5 on Beauty, by effectively modeling user behavioral uncertainty and enhancing recommendation diversity.

Key insights

  • Diffusion models overcome posterior collapse and model collapse issues prevalent in VAEs and GANs, enabling more robust generative modeling of sequential user behavior.
  • Noising only the target item while preserving the historical sequence context is crucial for effective sequential recommendation with diffusion models, preventing corruption of contextual information.
  • Efficient training (single-step sampling) and inference (direct prediction with multi-seed averaging) strategies are critical for practical deployment of diffusion models in large-scale recommender systems.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.