Research map / Vision-language and multimodal / Human and embodied modeling
$G^2$-Reader: Dual Evolving Graphs for Multimodal Document QA
Yaxin Du, Junru Song, Yifan Zhou, Cheng Wang, Jiahao Gu, Zimeng Chen, Menglan Chen, Wen Yao, Yang Yang, Ying Wen, Siheng Chen
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 multimodal (visual) evidence to retrieval and generation (ablation) | G2-Reader vanilla baseline · VisDoMBench 250-sample subset | average accuracy: 54.8 -> 63.2 | Improved |
| add Planning Graph on top of multimodal baseline (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 63.2 -> 64.4 | Improved |
| add Content Graph on top of multimodal baseline (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 63.2 -> 63.6 | Improved |
| combine Content Graph and Planning Graph (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 63.2 -> 68.0 | Improved |
| increase Planning Graph refinement rounds from 0 to 3 (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 64.8 -> 68.0 | Improved |
| increase refinement rounds beyond 3 (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 68.0 -> 64.4 | Hurt |
| evolve Content Graph for 3 iterations instead of 0 (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 61.8 -> 68.0 | Improved |
| evolve Content Graph for 5 iterations instead of 3 (ablation) | G2-Reader · VisDoMBench 250-sample subset | average accuracy: 68.0 -> 63.6 | Hurt |
| replace VLM-based evolution with rule-based lite update (ablation) | G2-Reader (T=3) · VisDoMBench 250-sample subset | average accuracy: 68.0 -> 67.0 | No clear effect |
| add structured subgraph readout (ablation) | G2-Reader (T=3) · VisDoMBench 250-sample subset | average accuracy: 66.0 -> 68.0 | Improved |
| use G2-Reader with Qwen3-VL-32B instead of prompting the model directly | Qwen3-VL-32B · VisDoMBench full | average accuracy: 29.90 -> 66.21 | Improved |
| use G2-Reader at k=5 versus vanilla RAG at k=20 | retrieval · recall | recall: 86.2 -> 86.3 | Improved |
Closest papers by meaning
- Docopilot: Improving Multimodal Models for Document-Level Understanding (CVPR 2025 · similarity 0.88)
- TopoGraphRAG-Bench: Evaluating Multimodal GraphRAG on Layout-Grounded Evidence Reasoning (NeurIPS 2026 · similarity 0.88)
- Benchmarking Retrieval-Augmented Multimomal Generation for Document Question Answering (NeurIPS 2025 · similarity 0.88)
- Knowledge Graph Prompting for Multi-Document Question Answering (AAAI 2024 · similarity 0.88)
- SPIQA: A Dataset for Multimodal Question Answering on Scientific Papers (NeurIPS 2024 · similarity 0.87)
- Resolving Evidence Sparsity: Agentic Context Engineering for Long-Document Understanding (CVPR 2026 · similarity 0.87)
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.
