Reproduce “LGMRec: Local and Global Graph Learning for Multimodal 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
Zhiqiang Guo, Jianjun Li, Guohui Li, Chaoyang Wang, Si Shi, Bin Ruan
The LGMRec multimodal recommender system decouples collaborative and multimodal signals for learning local user interests and incorporates a hypergraph-based module to capture global, modality-aware interests. This approach achieved up to 12.98% improvement in Recall@10 over prior methods on Amazon datasets, demonstrating enhanced accuracy and robustness in sparse interaction environments.
Key insights
- Decoupling collaborative and multimodal signals for local interest learning resolves training conflicts and leads to more stable and accurate user interest representations.
- Capturing modality-aware global user interests through hypergraph learning provides a robust way to enrich interest modeling, especially when local interaction data is sparse.
- Cross-modal contrastive learning is effective in aligning and making global embeddings consistent across different item modalities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.