MedRealMM
收藏资源简介:
MedRealMM是由京东健康国际有限公司牵头构建的大规模、真实世界中文在线医疗咨询多模态评测基准。该数据集包含5,620例源自全国性互联网医院平台的实际医患交互案例,覆盖64个临床科室,其核心特色在于保留了患者上传的医疗图像与文本对话交织的多模态语境。数据集通过创新的多模态临床挑战点提取框架,从真实咨询轨迹中识别出需要实质性临床推理的关键时刻,并将其转化为标准化的下一轮响应生成任务。该基准旨在精准评估大语言模型在融合视觉与文本信息、遵循临床安全规范、生成开放式诊疗建议等方面的综合能力,为解决人工智能在真实在线医疗场景中可靠性验证的核心瓶颈提供了重要工具。
MedRealMM is a large-scale, real-world Chinese online medical consultation multimodal evaluation benchmark led by JD Health International Inc. This dataset comprises 5,620 real doctor-patient interaction cases sourced from national internet hospital platforms, spanning 64 clinical departments. Its core characteristic is the retention of multimodal context that interleaves medical images uploaded by patients with text-based dialogues. Leveraging an innovative multimodal clinical challenge extraction framework, the dataset identifies critical moments requiring substantive clinical reasoning from authentic consultation trajectories, and converts these into standardized next-round response generation tasks. This benchmark is designed to accurately evaluate the comprehensive capabilities of Large Language Models (LLMs) in fusing visual and textual information, adhering to clinical safety regulations, and generating open-ended diagnostic and treatment recommendations, thereby providing a critical tool to address the core bottleneck of reliability verification for artificial intelligence in real-world online medical scenarios.
数据集概述:MedRealMM
- 许可证:Apache-2.0
该数据集页面未提供任务类型、数据规模、数据格式或具体用途等详细信息。仅标注了使用 Apache-2.0 开源许可证,允许用户自由使用、修改和分发。
- 1MedRealMM: A Real-World Multimodal Benchmark for Chinese Online Medical Consultation京东健康国际有限公司; 上海交通大学; 新加坡国立大学; 北卡罗来纳大学教堂山分校; 宾夕法尼亚大学 · 2026年



