Explainable AI for suicide risk detection: gender- and age-specific patterns from real-time crisis chats

Key Findings

• — Analysed 17,564 Hebrew-language crisis-chat sessions, of which 3,097 were classified as suicide-risk cases
• — A theory-driven lexicon of 20 constructs (e.g. hopelessness, loneliness, self-harm), derived from the Interpersonal Theory of Suicide, the Suicide Crisis Syndrome, and the Columbia framework, served as interpretable features
• — Stratified analyses revealed gender- and age-specific patterns: loneliness was a consistent predictor for women, thwarted belongingness was salient for men, and hopelessness and prior attempts predicted risk across groups
• — The approach prioritises explainability so that individual linguistic risk factors, rather than opaque scores, drive detection

Methodology Notes

Retrospective NLP analysis of 17,564 anonymized text-based crisis-chat sessions (Sahar; Hebrew chats only, Arabic-language chats excluded). Explainable-AI approach: a 20-construct psychological lexicon plus stratified logistic regression by gender and age; outcome is explicit suicidal ideation. The underlying dataset is not public (confidentiality agreement). Published 18 December 2025; a PMC mirror exists (PMC12756489).

Sources