Human-AI-Dialogue-Suicide-Risk-Dataset

Human-AI-Dialogue-Suicide-Risk-Dataset This dataset provides 4,040 annotated human-AI dialogue samples focused on mental health and suicide risk assessment. The detailed information of the dataset is available at DOI: 10.1016/j.ipm.2026.104787. If you use this dataset, please cite the following paper: Chen, X., Li, Q., Jiang, Y., Liu, M., Yang, L., Jing, X., Wei, Z., & Cao, L. (2026). PsychoDial: A framework for building a suicide risk dialogue dataset in human-AI conversations. Information Processing & Management, 63(7), 104787. DOI: 10.1016/j.ipm.2026.104787. 📌 Overview This dataset contains 4,040 human-AI dialogue samples focused on mental health, specifically related to suicide risk assessment, depression, and emotional support. Each entry consists of a multi-turn conversation between a user expressing distress and an AI providing empathetic responses. The dataset is annotated with specific risk labels to facilitate research in suicide risk detection and mental health analysis. 📂 Data Structure The dataset is provided in excel format and contains the following columns: index: A unique numerical identifier for each dialogue entry. dialogue: The full text of the conversation. The format follows a User: [Text] \n AI: [Text] structure, capturing the exchange of thoughts, feelings, and supportive responses. post_risk: The annotation label indicating the category of suicide risk or the nature of the distress expressed in the dialogue. 🏷️ Risk Labels (Classes) The dialogues are classified into the following categories based on the user's expression of risk: ideation: The user expresses thoughts of dying, wishing to be dead, or general suicidal thoughts without a specific, immediate plan. behavior: The user describes specific actions, preparations, or behavioral patterns (e.g., self-isolation, giving away possessions, acquiring means) that suggest an increased risk or intent. indicator: The user exhibits signs or symptoms strongly associated with severe distress or potential risk, such as feelings of hopelessness, worthlessness, or being a burden to others. attempt: The user discusses past suicide attempts, imminent plans, or engaging in self-harm with the intent to end their life. safe: Dialogues where the user discusses mental health but does not exhibit immediate suicide risk, or discusses recovery and coping strategies. 🚀 Potential Applications This dataset is designed for researchers and developers working in: Natural Language Processing (NLP) for mental health. Suicide Risk Detection: Training models to classify the severity of risk in text. Sentiment Analysis: Understanding emotional distress in conversational contexts. Conversational AI: Training empathetic chatbots for crisis intervention and emotional support. ⚠️ Ethical Considerations & Disclaimer Sensitive Content: This dataset contains textual descriptions of severe mental distress, suicide ideation, and self-harm. Please approach the data with care. Research Use Only: This dataset is intended for academic and research purposes only. It is not a substitute for professional clinical diagnosis or treatment. Bias & Safety: Users of this dataset should be aware that AI models trained on this data should be rigorously tested for safety and bias before deployment in any real-world setting.