Benchmark RadarMKG-RAG-Bench
MKG-RAG-Bench is a benchmark for retrieval in multimodal knowledge graph-augmented generation, constructed from two knowledge graphs (general and medical) with QA datasets supporting controlled evaluation of retrieval and generation. It uses an LLM-based curation pipeline for structurally grounded queries.
- Released
- 2026-06-24
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Isolates retrieval as a first-class evaluation target in MKG-RAG, addressing the challenge of heterogeneous multimodal knowledge. It provides a foundation for diagnosing retrieval limitations and improving end-to-end RAG systems.
Motivation
Retrieval-augmented generation (RAG) over knowledge graphs has emerged as a promising approach for grounding large language models, yet existing benchmarks largely overlook the challenges of retrieval in multimodal knowledge graph RAG (MKG-RAG).
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.