Towards Emotion-and Time-Aware Classification of Tweets to Assist Human Moderation for Suicide Prevention

Manual moderation of social media content for suicide prevention is a critical but overwhelming task for human moderators. We propose an emotion- and time-aware classification system designed to assist human moderators by intelligently prioritizing tweets that require immediate attention for suicide prevention. Our approach recognizes that both emotional content and temporal patterns are crucial factors in identifying urgent cases that need human intervention. The system incorporates sophisticated emotion detection mechanisms that can identify subtle emotional cues indicative of suicide risk, while also considering temporal factors such as posting frequency, time of day, and temporal clustering of concerning posts. We develop a multi-modal classification framework that combines textual analysis with temporal behavioral patterns to create a comprehensive risk assessment. The system is designed to work as a decision support tool for human moderators, providing risk scores and explanations to help them prioritize their limited time and attention. Through evaluation on real-world social media datasets and collaboration with mental health professionals, we demonstrate that our emotion- and time-aware approach significantly improves the efficiency and effectiveness of human moderation for suicide prevention.