PubMedQA-Search

In real application scenarios, users often do not know the reference knowledge corresponding to their questions; they simply want to ask a question and receive an answer. The PubMedQA-Search dataset is designed to simulate and test EBM scenarios. We constructed two sub-datasets: PubMedQA-Search-Short and PubMedQA-Search-Long. The former uses Questions from the PubMedQA dataset as search queries, totaling 500 entries with an average length of 13.10 words. The latter uses Contexts and Long Answers from the PubMedQA dataset as search queries, totaling 2189 entries with an average length of 54.94 words. Both are labeled with the corresponding PMID.