RAGRouter-Bench: A Multi-Domain Dataset for Adaptive RAG Routing

RAGRouter-Bench is a dataset and benchmark designed for adaptive Retrieval-Augmented Generation routing. It contains 7,727 queries and 21,460 documents across four domains: Wikipedia, Literature, Legal, and Medical. The dataset was created by Chaplain0908 and last updated on January 23, 2026.

Use Cases

• Benchmarking RAG system performance based on the described multi-domain query-document corpus.
• Evaluating query-corpus compatibility for routing decisions based on the three query types: Factual, Reasoning, and Summary.
• Training adaptive retrieval models based on the dataset's structured queries and documents.
• Comparing retrieval strategies across the four specified domains: Wikipedia, Literature, Legal, and Medical.

Strengths

• Contains 7,727 queries and 21,460 documents, providing substantial material for evaluation.
• Covers four distinct domains, likely offering diverse linguistic and structural contexts.
• Includes three specific query types, which may allow for nuanced performance analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Chaplain0908 on Hugging Face.

Collection Method

Constructed from existing datasets: MuSiQue, QuALITY, UltraDomain, and GraphRAGBench.

Time Range

null

Freshness

Last updated 2026-01 23 07:37:45; freshness should be verified.

Geography

null