The algorithm that Prevents Suicide
Imagine a system that learns to detect a user's risk of suicide by using social media text. Some years ago, this would have been futuristic. Today, it's a reality, shown by several studies. "You can assess an individual's risk of having suicidal thoughts by looking at his social media output", Dirk Hovy says.
These methods, the outcome of an interdisciplinary effort, can complement traditional psychological procedures, namely impressions formed during psychologist sessions, and help assess patients on an ongoing basis. Existing studies, however, still need to model each mental condition in isolation, and do not take the patient's demographics into account. But conditions such as anxiety, depression, eating disorders, panic attacks, schizophrenia, bipolar disorder, and post-traumatic stress disorder often occur in clusters, and are highly correlated with age and gender.
In Multi-task Learning for Mental Health Conditions with Limited Social Media Data, a paper co-authored with Adrian Benton and Margaret Mitchell, Dirk Hovy shows how computers can help psychologists diagnose people's risk for mental health conditions. The authors created an algorithm to take into account not only patients' written texts, but also demographic variables and potential other mental health conditions. "This is the same thing a psychologist does: if the patient is, for example, a man under 30, it guides them to assess the symptoms in a different way than if it were a woman over 60. We showed that a model that does the same can predict the risk for several mental health conditions much better than prior systems". A difference in performance that translates into 120 additional correctly predicted suicide attempts. "This is a promising new direction in automated assessment of mental health and suicide risk, with possible application to the clinical domain". A direction, of course, that needs to include psychologists, computer scientists, and most of all patients.
Read more about this topic:
Riccardo Zecchina. Teaching Machines How to Learn to Improve Business and Life
Carlo Baldassi. Learning Is a Quantum Question
Daniele Durante. How to Study Dynamic Networks
Antonio Lijoi, Igor Pruenster. Algorithms for Analysis of Complicated Phenomena
Alessia Melegaro. The Network Modeling Chip that Fights Influenza
Raffaella Piccarreta, Marco Bonetti. A (Statistical) Model for Life