BEncoderRT/User_Intent_Risk_Triage

这是一个高质量的中文数据集，专门设计用于训练语言模型在对话上下文中进行心理健康风险评估。它支持单轮用户陈述和长多轮对话历史。数据集包含真实的中文心理困扰表达，包括间接和微妙的自杀意念，以及长自然多轮对话中的升级风险信号。数据集还包含助理在对话历史中展示的安全、同理心的危机干预实践。标签强调伦理分级：当存在自杀念头时，升级并提供资源。数据集涵盖从低（轻度困扰）到中/高（明显抑郁+被动/主动自杀意念）的各种风险级别。

This is a high-quality Chinese dataset specifically designed for training language models to perform mental health risk assessment in conversational contexts. It supports both single-turn user statements and long multi-turn conversation histories. The dataset contains realistic Chinese expressions of mental distress, including indirect and subtle suicidal ideation, as well as long, naturalistic multi-turn conversations with escalating risk signals. The dataset also includes assistant responses in the conversation history that demonstrate safe, empathetic crisis intervention practices. Labels emphasize ethical triage: escalate and provide resources when suicidal thoughts are present. The dataset covers a range of risk levels from low (mild distress) to medium/high (clear depression + passive/active suicidal ideation).