Reproduce “Gaussian Mixture Flow Matching with Domain Alignment for Multi-Domain Sequential Recommendation” 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
Xiaoxin Ye, Chengkai Huang, Hongtao Huang, Lina Yao
GMFlowRec introduces a generative framework for Multi-Domain Sequential Recommendation, unifying Gaussian Mixture Flow Matching with domain-aware representation learning to model complex user behaviors across multiple domains. It achieves state-of-the-art performance on JD and Amazon datasets, demonstrating up to 44.37% NDCG@5 improvement in challenging scenarios while maintaining computational efficiency.
Key insights
- Gaussian Mixture Flow Matching can effectively capture the multi-modal nature of user preferences and uncertain next-item transitions in multi-domain sequences, overcoming limitations of uni-modal priors.
- Disentangling global and local user intents through a dual-masked Transformer within a shared backbone is an efficient and scalable approach for MDSR, avoiding the complexity of multiple per-domain networks.
- Aligning the generative process with domain-aware user history via a domain-aligned prior is crucial for improving performance, especially in scenarios with diverse target domains or limited data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.