Using natural language processing to improve mental health support
Using natural language processing to improve mental health support Heading link
May is Mental Health Awareness Month, and UIC Engineering research aims to help mental health professionals identify those who are most vulnerable with better support and screening tools.
Those who suffer from psychotic disorders such as schizophrenia and bipolar disorder face a higher risk of suicide. These individuals also experience increased suicidal ideation, a higher likelihood of progression to a suicide attempt, and more severe suicide attempts compared with the general population. Despite this, those in the field say more research needs to be done to prevent suicide in this group.
Assistant Professor Natalie Parde is developing ways to automatically detect linguistic and verbal features of psychotic disorders, with the goal of improving automated mental health support and screening. In the long term, her team hopes that this work could be leveraged to reduce downstream escalation (e.g., through lower rates of suicide) for people with these disorders.
Clinical psychologists at the University of California – San Diego, the University of Texas – Dallas, and the University of Miami are collecting data via speech recordings with individuals with mental health issues. Over 300 samples from consensual patient-doctor recordings have been shared to date with Parde, with additional data coming.
Using advanced methods of natural language processing – a branch of artificial intelligence that focuses on the automated interpretation and generation of human language – Parde then analyzes the samples of natural language data.
“We are transcribing these speech recordings to enable easier natural language processing, and then extracting specialized features from these transcripts, encoding different and potentially valuable characteristics of the spoken language,” Parde said.
Preliminary data indicates that there is a link, over time, in how cognitive biases may be relevant to suicidal thinking and behavior in people with psychotic disorders. These biases include the ways patients perceive others, such as the risk of threat from others, or the value of relationships.
The data is fed into machine learning algorithms with the end goal of automatically detecting mental health issues from the language and phrasing used to help mental health providers detect suicide ideation in these patients.
The four-year study involves recordings of those with psychotic disorders, as well as a control group of patients. This allows the researchers to empirically analyze the language differences between members of those groups. Parde said a barrier to this type of work is the difficulty in obtaining data sets for these tasks.
“We plan to make our data and results publicly available so others can build upon this work and develop models of their own,” Parde said.
Her $117K grant is part of a $2.34M National Institutes of Health project under Colin A. Depp at the University of California – San Diego, Social Cognition and Suicide in Psychotic Disorders.