1,000 real consumer health questions paired with an expert-condensed one-sentence summary, from the ACL 2019 paper that introduced medical question summarisation as a task.
unassessed
| Category | generation |
|---|---|
| Subcategory | abstractive summarisation of long consumer health questions into a single focused question |
| Page status | active |
| Metric | ROUGE-1/2/L, BLEU, BERTScore and BLEURT (harness-dependent) |
| Direction | higher_is_better |
| Unit | % |
| Dataset size | 1000 |
| Dataset licence | Stated two ways: the original GitHub repository's README states CC BY 4.0 explicitly; the `bigbio/meqsum` mirror on Hugging Face lists its licence as "unknown." |
| Publisher | U.S. National Library of Medicine (NIH) |
MeQSum tests whether a model can compress a long, often rambling real-world consumer health question -- the kind a patient submits to a health Q&A service, with extra context, background details and multiple sub-questions folded in -- into a single, focused question that preserves the core information need. The paper's own motivation is practical: overly long questions increase false positives in downstream answer retrieval, so summarising the question first, before trying to answer it, is framed as a way to make question-answering systems more accurate. Inputs and outputs are both English text; there is no multiple-choice or classification step, only free-text generation.
Single-document abstractive summarisation: the model reads one consumer health question (CHQ) and generates a single condensed summary question, scored against one expert-written reference summary per item.
No model card in ModelSpec reports this benchmark yet.