Identification of Imminent Suicide Risk Among Young Adults using Text Messages
summary
Study recruits 26 participants and extracts ~500,000 outgoing texts, with ~140,000 messages labeled as depression or suicidality, aggregated into 528 depression days and 685 suicidality days. Messages convert to tf-idf and feed a tf-idf + 5-layer DNN (1,000 nodes/layer, dropout, stratified cross‑validation), which flags suicidal days correctly 81% of the time but produces a 44% false alarm rate; higher daily message counts reduce misclassification, and informal, personalized language limits generalization.
AI-written summary · about
chapters (13)
- 0:00Introduction to study
- 1:09Current challenges in suicide risk prediction
- 2:15Recruitment and screening process
- 4:44Lab session and data collection
- 6:07Identifying mental health episodes
- 7:31Text message data preparation
- 8:15Modeling approach and results
- 9:18Performance and limitations
- 11:05Future directions and improvements
- 12:32Potential clinical applications
- 13:35Ethical and legal considerations
- 15:01Ongoing challenges and research
- 16:26Q&A session
related talks
32mAnalyzing social media for suicide risk using natural language processing
20mAugmenting Mental Health Care in the Digital Age: Machine Learning as a Therapist Assistant
27mHARSHA URLAM & SWETA SINGH at Cypher 2017
40mBringing mental health data to doctors
06mLeveraging Collaborative-Filtering for Personalized Behavior Modeling: A Case Study of ...
37mDetecting offensive messages using Deep Learning: A micro-service based approach
32mTriaging Feedback Form Data
30mIncreasing the trading prediction by mining aggregated human texting messages