HKAI-Sci/qkg_qa_dataset

`qkg_qa_dataset` 是QKG论文第4.1节中描述的经过整理的QKG评估数据集。它包含`2,788`个为基于图的评估而选择的多项选择生物医学QA样本，以及：原始问题和答案选择、PrimeKG和UMLS支持的实体链接结果、QKG子图证据和关系适用性注释、结构化患者特征提取、模型评估输出（`v1_eval`和`v3_eval`）。发布的文件是`top_samples_filtered.jsonl`。该数据集旨在用于重现QKG论文评估数据集、研究基于图的生物医学QA与患者感知关系约束、分析整理的QA样本如何连接到PrimeKG子图和QKG证据。

`qkg_qa_dataset` is the curated QKG evaluation dataset described in Section 4.1 of the QKG paper. It contains `2,788` multiple-choice biomedical QA samples selected for graph-grounded evaluation, together with: the original question and answer choices, entity linking results over PrimeKG and UMLS-backed alignments, QKG subgraph evidence and relation applicability annotations, structured patient-character extraction, model evaluation outputs (`v1_eval` and `v3_eval`). The published file is `top_samples_filtered.jsonl`. This artifact is released for reproducing the QKG paper evaluation dataset, studying graph-grounded biomedical QA with patient-aware relation constraints, analyzing how curated QA samples connect to PrimeKG subgraphs and QKG evidence.