Research map / Graph learning / Graph self-supervised learning
Multimodal Graph Representation Learning with Dynamic Information Pathways
Xiaobin Hong, Mingkai Lin, Xiaoli Wang, Chaoqun Wang, Wenzhong Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add global (pseudo-node) message passing (ablation) | DiP · Amazon-Sports link prediction | MRR: 29.48 -> 34.26 | Improved |
| add local message passing (ablation) | DiP · Amazon-Sports link prediction | Hit@10: 61.34 -> 72.54 | Improved |
| add cross-modal pseudo-node interaction P(v)<->P(t) (ablation) | DiP · Ele-Fashion | accuracy (%): 82.47 -> 87.88 | Improved |
| add visual pseudo nodes (ablation) | DiP · Ele-Fashion | accuracy (%): 84.62 -> 87.88 | Improved |
| add textual pseudo nodes (ablation) | DiP · Ele-Fashion | accuracy (%): 85.34 -> 87.88 | Improved |
| add pseudo nodes (G2P/P2G) to propagate messages vs static GNNs when depth grows | DiP · Node classification, message passing hops 1-10 | accuracy | Improved |
| use DiP pseudo-node message passing instead of dense/static multimodal GNNs | MMGCN, MGAT · Amazon-Sports | memory per epoch (MB): 2570.52 -> 660.24 | Improved |
| use DiP instead of GCN/SAGE in compute time | GCN · Ele-Fashion | time per epoch (s): 0.253 -> 0.531 | Hurt |
| use DiP instead of MGAT with DINOv2/T5 encoders | MGAT · Goodreads-LP (DINOv2/T5) | MRR: 74.89 -> 44.76 | Hurt |
| vary number of visual/textual pseudo nodes (16 to 512) (ablation) | DiP · Ele-Fashion / Amazon-Sports | accuracy / MRR | Mixed |
| use DiP over best baseline across encoder choices | DiP · Goodreads-NC (CLIP) | accuracy (%): 83.30 -> 85.37 | Improved |
Closest papers by meaning
- Graph4MM: Weaving Multimodal Learning with Structural Information (ICML 2025 · similarity 0.88)
- Mosaic of Modalities: A Comprehensive Benchmark for Multimodal Graph Learning (CVPR 2025 · similarity 0.88)
- Mario: Multimodal Graph Reasoning with Large Language Models (CVPR 2026 · similarity 0.86)
- Multimodal Graph Neural Architecture Search under Distribution Shifts (AAAI 2024 · similarity 0.85)
- Towards Dynamic Message Passing on Graphs (NeurIPS 2024 · similarity 0.85)
- GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning (CVPR 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
