MedNLI (Medical Natural Language Inference)

MedNLI 数据集由医生从 MIMIC-III 临床笔记的过去病史部分开发的句子对组成，注释为绝对正确、可能正确和绝对错误。该数据集包含 11,232 个训练实例、1,395 个开发实例和 1,422 个测试实例。这提供了基于患者病史的自然语言推理任务 (NLI)。

The MedNLI dataset consists of sentence pairs developed by physicians from the past medical history sections of MIMIC-III clinical notes, annotated as definitely correct, probably correct, and definitely incorrect. It contains 11,232 training instances, 1,395 development instances, and 1,422 test instances. This provides a natural language inference (NLI) task based on patient medical histories.

搜集汇总

背景与挑战

背景概述

MedNLI是一个医疗自然语言推理数据集，基于MIMIC-III临床笔记的过去病史构建，包含医生标注的句子对，分为绝对正确、可能正确和绝对错误三类。该数据集规模为11,232个训练实例、1,395个开发实例和1,422个测试实例，旨在支持基于患者病史的自然语言推理任务研究。