Reproduce “Robust Multimodal Recommendation via Graph Retrieval-Enhanced Modality Completion” 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
Yuan Li, Jun Hu, Jiaxin Jiang, Bryan Hooi, Bingsheng He
Researchers at the National University of Singapore developed GRE-MC, a framework that addresses missing modalities in multimodal recommendation systems by retrieving semantically relevant subgraphs and jointly encoding them with a graph transformer. The system achieved 2.3% to 7.1% superior performance in recommendation metrics over existing missing-modality-aware baselines across various datasets.
Key insights
- Leveraging globally available, semantically relevant context from the entire graph, beyond just local neighborhoods, provides stronger cues for reconstructing missing modalities.
- Graph Transformers can effectively encode a query node with a potentially disconnected, retrieved subgraph, enabling the capture of crucial long-range dependencies for robust completion.
- Regularizing latent embeddings through a sparse-routing codebook enhances the model's robustness and generalization capability when dealing with partial information.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.