TigerResearch/MedCT-clinical-notes

---
license: apache-2.0
---

A real-world clinical dataset in Chinese, for a variety of LLM-based applications:
* For the NER and NEL tasks, 7.4K real-world clinical notes in Chinese (medct_ner_notes.csv), and 61K entity mention annotations per MedCT graph (medct_ner_annotations.csv).
* For the search task, 20 clinical queries (medct_search_queries.csv), and 2K discharge notes with relevance annotations (medct_search_notes.csv).
* For the clinical notes summarization task, 91 raw discharge notes with summary by human, LLM and MedCT-augmented generations (medct_summary_notes.csv), with preference Likert-scale annotated by human physicians (medct_summary_scores.csv).

许可证：Apache-2.0

本数据集为面向各类基于大语言模型（LLM）应用的中文真实临床数据集：
* 针对命名实体识别（NER）与实体链接（NEL）任务，包含7400条中文真实临床病历文本（medct_ner_notes.csv），以及基于MedCT图谱的61000条实体提及标注数据（medct_ner_annotations.csv）。
* 针对搜索任务，包含20条临床查询语句（medct_search_queries.csv），以及带有相关性标注的2000份出院病历文本（medct_search_notes.csv）。
* 针对临床病历摘要生成任务，包含91份原始出院病历文本，分别配有人工、大语言模型（LLM）以及MedCT增强生成的摘要（medct_summary_notes.csv），并附带由临床医师采用李克特（Likert）量表进行偏好性标注的评分数据（medct_summary_scores.csv）。