Document Question Answering on M3DocVQA and FRAMES (Average)
19EMHiKEY
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| HiKEYMethod Category=Multimodal GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 19 | 22.7 | 25 | 19.7 | |
| MultiDocFusionMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 15.4 | 18.4 | 20.6 | 16.1 | |
| HopRAGMethod Category=Text-based GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 15.3 | 18.2 | 20.5 | 16 | |
| SimpleDocMethod Category=Multimodal GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 15.3 | 17.8 | 20 | 15.1 | |
| RAPTORMethod Category=Text-based GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 15 | 17.9 | 20.1 | 15.7 | |
| MoLoRAGMethod Category=Multimodal GraphRAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 15 | 17.4 | 19.8 | 14.9 | |
| VDocRAGMethod Category=Page-level multimodal RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 14.5 | 16.8 | 19.1 | 14.4 | |
| M3DocRAGMethod Category=Page-level multimodal RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 14.2 | 16.5 | 18.6 | 14 | |
| PageMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 14.1 | 16.9 | 18.9 | 14.8 | |
| LumberChunkerMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 14.1 | 16.8 | 19.1 | 14.9 | |
| Meta ChunkerMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 14.1 | 16.8 | 19 | 14.8 | |
| Structural chunkingMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 13.5 | 16.1 | 18.3 | 14.3 | |
| Length chunkingMethod Category=Text chunk-based RAG, Input budget=16K, LVLM Reader=same LVLM Reader2026.05 | 12.2 | 14.8 | 17.2 | 13.6 |