Analyzing social media for suicide risk using natural language processing
summary
Researchers apply NLP and machine learning to social media and clinical text, build a donation platform (Our.Data.Helps.org) and a university mirror (UMD.OurDataHelps.org), and run a shared task inside a secure data enclave on AWS. Models (topic models, a three-level Hierarchical Attention Network, and tiered multi-armed bandits for prioritization) show much higher detection rates than clinicians (roughly 40–60% true positives vs 4–6%), improve baselines in the shared task, and in simulation find more of 44 severe cases within a three-hour clinician budget.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:39Problems clinicians face
- 3:55Metaanalysis
- 8:03Analysis
- 9:55Challenges
- 10:16Clinical White Space
- 13:09Making Progress
- 14:42Secure Data Enclave
- 16:23How it works
- 18:17What is the problem
- 19:36Two shifts
- 22:12Technical approach
- 23:59Results
- 25:41Multiarmed bandits
- 27:12A prioritization framework
- 29:26Summary
related talks
20mAugmenting Mental Health Care in the Digital Age: Machine Learning as a Therapist Assistant
20mIdentification of Imminent Suicide Risk Among Young Adults using Text Messages
33mLessons Learned Applying Large Language Models in Healthcare
25mDeep learning for disaster management and response
57mHow Machine Learning and AI Can Support the Fight Against COVID-19
26mNatural Language Processing for Healthcare: Current Trends...
23mLeveraging Generative AI and Transformers in Healthcare
40mBringing mental health data to doctors