Document Question Answering on M3DocVQA
61Exact MatchBookRAG
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BookRAGBaseline Type=Our proposed2025.12 | 61 | 66.2 | — | — | — | |
| GraphRankerBaseline Type=Layout segmented RAG2025.12 | 43 | 47.8 | — | — | — | |
| DocETLBaseline Type=Layout segmented RAG2025.12 | 40.9 | 43.3 | — | — | — | |
| Layout + VanillaBaseline Type=Conventional RAG2025.12 | 36.9 | 40.2 | — | — | — | |
| Vanilla RAGBaseline Type=Conventional RAG2025.12 | 36.5 | 40.2 | — | — | — | |
| BM25Baseline Type=Conventional RAG2025.12 | 34.6 | 37.8 | — | — | — | |
| RAPTORBaseline Type=Graph-based RAG2025.12 | 34.3 | 37.3 | — | — | — | |
| Tree-TraverseBaseline Type=Layout segmented RAG2025.12 | 33.3 | 36.2 | — | — | — | |
| HiKEYMethod Category=Multimodal GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 27.5 | — | 30.7 | 32.2 | 23.9 | |
| SimpleDocMethod Category=Multimodal GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 25.6 | — | 28.7 | 30.1 | 22.5 | |
| MoLoRAGMethod Category=Multimodal GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 25.2 | — | 28.2 | 30 | 22.4 | |
| MM-VanillaBaseline Type=Layout segmented RAG2025.12 | 25.1 | 27.7 | — | — | — | |
| VDocRAGMethod Category=Page-level multimodal RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 24.4 | — | 27.4 | 28.8 | 21.6 | |
| M3DocRAGMethod Category=Page-level multimodal RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 24.1 | — | 27 | 28.2 | 21.3 | |
| GraphRAG-LocalBaseline Type=Graph-based RAG2025.12 | 23.7 | 25.6 | — | — | — | |
| MultiDocFusionMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 23 | — | 25.9 | 27.3 | 20.6 | |
| HopRAGMethod Category=Text-based GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 22.8 | — | 25.7 | 27.1 | 20.5 | |
| RAPTORMethod Category=Text-based GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 22.5 | — | 25.3 | 26.7 | 20.2 | |
| LumberChunkerMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 21.4 | — | 24.2 | 25.6 | 19.5 | |
| Meta ChunkerMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 21.3 | — | 24.1 | 25.5 | 19.4 | |
| PageMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 21.1 | — | 24 | 25 | 19 | |
| Structural chunkingMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 20.6 | — | 23.3 | 24.6 | 18.8 | |
| GraphRAG-GlobalBaseline Type=Graph-based RAG2025.12 | 20.2 | 22 | — | — | — | |
| Length chunkingMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 17.5 | — | 19.9 | 21.4 | 16.7 |