ReDiX/

ReDiX-Benchmark-Medical-ita

A medical-domain benchmark dataset for evaluating retrieval, reranking, and RAG systems under low lexical overlap and high semantic difficulty. ⚠️ Designed to penalize shallow matching.High scores from lexical methods (e.g., BM25) may indicate shortcut exploitation, not real understanding.

This dataset is built to:

reduce lexical similarity between queries and relevant content
increase semantic diversity across… See the full description on the dataset page: https://huggingface.co/datasets/ReDiX/ReDiX-Benchmark-Medical-ita.