MENTALBENCH
收藏资源简介:
MENTALBENCH是由韩国科学技术院等机构联合开发的精神病诊断评估基准,其核心为专家构建的知识图谱MENTALKG,涵盖23种精神障碍的DSM-5标准。数据集包含24,750条合成临床案例,通过规则增强生成并经过精神病学家审核,确保临床合理性。案例设计涵盖从结构化医疗记录到不完整患者自述的信息谱系,以及从单一病症到复杂鉴别诊断的场景,旨在系统性评估模型在模糊边界下的诊断决策能力。该数据集填补了现有基准在精神病学逻辑严谨性和诊断过程模拟方面的空白,为AI辅助精神健康分析提供标准化测试平台。
MENTALBENCH is a psychiatric diagnosis evaluation benchmark co-developed by institutions including the Korea Advanced Institute of Science and Technology (KAIST). Its core is the expert-constructed knowledge graph MENTALKG, which covers the DSM-5 diagnostic criteria for 23 types of mental disorders. The dataset contains 24,750 synthetic clinical cases, which are generated via rule-augmented generation and reviewed by psychiatrists to ensure clinical plausibility. The case design spans the full information spectrum from structured medical records to incomplete patient self-reports, as well as scenarios ranging from single-condition presentations to complex differential diagnoses. It aims to systematically evaluate the diagnostic decision-making capabilities of models under ambiguous diagnostic boundaries. This dataset fills the gap in existing benchmarks regarding the logical rigor of psychiatry and the simulation of diagnostic processes, providing a standardized testing platform for AI-assisted mental health analysis.



