[arXiv]score: 0.24

RAG Performance Varies Significantly by Query Type and Language

August 18, 2026

Evaluating RAG architectures for Bengali agricultural advisory reveals that BM25 outperforms dense retrieval for colloquial queries (R@10 = 0.506), while Hybrid RRF achieves the highest overall R@10 of 0.539. Dense retrieval performance drops from 0.970 on formal queries to 0.093 on colloquial farmer queries.

HOW THIS AFFECTS YOU

●

builderYou should consider hybrid retrieval approaches rather than relying solely on dense embeddings for colloquial low-resource languages.

●

researcherThis highlights the necessity of testing retrieval performance across specific linguistic registers rather than using aggregate scores.

read original ↗arxiv.org

DAILY DIGEST

you don't check 9 sources — we do. one email every morning, read in 2 min. free. unsubscribe anytime. privacy

← back to feed