Research map / Graph learning / Graph self-supervised learning
OpenMAG: A Comprehensive Benchmark for Multimodal-Attributed Graph
Chenxi Wan, Xunkai Li, Yilong Zuo, Haokun Deng, Sihan Li, Bowen Fan, Hongchao Qin, Rong-Hua Li, Guoren Wang
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 |
|---|---|---|---|
| use multimodal features instead of unimodal text or visual features (ablation) | MAG learning models · node classification and link prediction across datasets | accuracy / MRR | Improved |
| add graph neighbor aggregation (Graph-QFormer, 4 neighbors) to G2Image (ablation) | G2Image on SemArt · SemArt | DINO-Score: 50.86 -> 55.29 | Improved |
| increase aggregated neighbors from 4 to 8 (ablation) | G2Image on SemArt · SemArt | CLIP-Score: 70.21 -> 67.10 | Hurt |
| add graph neighbor aggregation (6 neighbors) to G2Image (ablation) | G2Image on Toys · Toys | DINO-Score: 28.13 -> 30.83 | Improved |
| fine-tune text and visual encoders end to end instead of freezing (ablation) | MLP on node classification · Movies and Grocery | test accuracy | Improved |
| use unified encoders (CLIP, Qwen2.5) instead of independent text plus image encoders (ablation) | MLP on modality retrieval · Flickr30k | retrieval Hits@K and MRR | Improved |
| use graph-enhanced models (LGMRec, DGF) instead of MLP and traditional GNNs | graph-based tasks · Movies, DY link prediction | accuracy / MRR: 62.57 -> 77.28 MRR (MLP to DGF on DY) | Improved |
| use MLLM-style generative masked-prediction model (UniGraph2) instead of discriminative models | graph-based tasks · Movies, Grocery, Toys | accuracy / NMI: 51.22 -> 46.78 acc (MLP to UniGraph2 on Movies) | Hurt |
| add graph structure aggregation to MLP for modality retrieval | MLP on modality retrieval · Toys | T2I MRR: 99.46 -> 94.20 (DGF) | Hurt |
| add LoRA fine-tuning to the LLM (ablation) | OPT on G2Text · Flickr30k | CIDEr: 41.51 -> 46.23 | Improved |
| add Adapter fine-tuning to the LLM (ablation) | OPT on G2Text · Flickr30k | CIDEr: 41.51 -> 34.95 | Hurt |
| add LoRA fine-tuning to the LLM (ablation) | LLaMA on G2Text · Flickr30k | CIDEr: 44.58 -> 47.76 | Improved |
Closest papers by meaning
- Mosaic of Modalities: A Comprehensive Benchmark for Multimodal Graph Learning (CVPR 2025 · similarity 0.91)
- Mario: Multimodal Graph Reasoning with Large Language Models (CVPR 2026 · similarity 0.89)
- GRAPHGPT-O: Synergistic Multimodal Comprehension and Generation on Graphs (CVPR 2025 · similarity 0.88)
- TEG-DB: A Comprehensive Dataset and Benchmark of Textual-Edge Graphs (NeurIPS 2024 · similarity 0.88)
- GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning (CVPR 2026 · similarity 0.88)
- InstructG2I: Synthesizing Images from Multimodal Attributed Graphs (NeurIPS 2024 · similarity 0.88)
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.
