Research map / Vision-language and multimodal / Visual question answering
M^3KG-RAG: Multi-hop Multimodal Knowledge Graph-enhanced Retrieval-Augmented Generation
Hyeongcheol Park, Jiyoung Seo, Jaewon Mun, Hogun Park, Wonmin Byeon, Sung June Kim, Hyeonsoo Im, JeungSub Lee, Sangpil Kim
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 modality-wise retrieval over the multi-hop MMKG (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 36.62 -> 40.91 | Improved |
| add GRASP on a text-only KG without modality-wise retrieval (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 36.62 -> 36.96 | No clear effect |
| combine modality-wise retrieval with GRASP (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 40.91 -> 44.67 | Improved |
| add GroundingDINO visual grounding to GRASP (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 40.91 -> 41.35 | Improved |
| add TAG audio grounding to GRASP (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 40.91 -> 41.70 | Improved |
| use both visual and audio grounding in GRASP (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 40.91 -> 42.96 | Improved |
| add LLM-based filtering stage to GRASP (ablation) | M3KG-RAG with Qwen2.5-Omni · VALOR | Model-as-Judge score: 42.96 -> 44.87 | Improved |
| LLM filter increases VRAM and latency (ablation) | GRASP · VALOR | GPU VRAM (GB): 24.2 -> 39.8 | Hurt |
| use Wikidata text KG with naive RAG instead of no retrieval | Qwen2.5-Omni · VALOR | Model-as-Judge score: 32.42 -> 30.28 | Hurt |
| use VAT-KG single-hop MMKG instead of no retrieval | Qwen2.5-Omni · VALOR | Model-as-Judge score: 32.42 -> 35.44 | Improved |
| set distance threshold tau too small or too large (ablation) | M3KG-RAG · VALOR | Model-as-Judge score | Hurt |
| raise GRASP presence threshold eta_av from 1.2 to 1.5 and 1.8 (ablation) | M3KG-RAG · VALOR | Model-as-Judge score | Hurt |
Closest papers by meaning
- Beyond Global Similarity: Multi-Conditional Retrieval for Fine-Grained Cross-Modal Understanding (CVPR 2026 · similarity 0.89)
- MC-Search: Evaluating and Enhancing Multimodal Agentic Search with Structured Long Reasoning Chains (ICLR 2026 · similarity 0.89)
- Re-ranking Reasoning Context with Tree Search Makes Large Vision-Language Models Stronger (ICML 2025 · similarity 0.88)
- Benchmarking Multimodal Retrieval Augmented Generation with Dynamic VQA Dataset and Self-adaptive Planning Agent (ICLR 2025 · similarity 0.88)
- MRAG-Bench: Vision-Centric Evaluation for Retrieval-Augmented Multimodal Models (ICLR 2025 · similarity 0.88)
- Document Haystacks: Vision-Language Reasoning Over Piles of 1000+ Documents (CVPR 2025 · 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.
