Innovation

Modeling and extraction of depression from social media

Project Leads: Randy Goebel, Osmar Zaïane, University of Alberta’s Faculty of Science


Is it possible to measure depression from the words of a tweet?


The advances of digital communications and technology provide a rich source for capturing and using key data for precision health purposes, including mental health.


Based on an ongoing project in collaboration between the University of Alberta, University of Ottawa and University of Montpellier, the research team seeks to extend the development of models on depression based on data from social media, including Facebook, Twitter and SMS.


These social media platforms provide data to apply machine learning to build models of depression, including the ability to calibrate the degree, intensity and change in depressive behaviour.


Goals

Integrated knowledge for a better diagnosis of depression

The team has engaged in two groups of clinicians connected with the cognitive and clinical physiology of mood, sentiment and depression. University of Alberta cognitive psychologists provide their expertise in the identification of mood and sentiment based on language analysis,
for example, to conduct a linguistic analysis that maps the language used in social media to standardized classifications of affective mood—like happy, sad or anxious, to name a few.


Linking this background with expertise in psychiatric clinical care, we can shed light on which individual attributes and indicators provide the most accurate assessment of both the existence and intensity of depression. The role of the data scientists is to integrate the background knowledge of psychologists and clinical psychiatrists together with a variety of data, and design and execute the appropriate experiments on choices of machine learning methods to build evaluable models of predicting depression from social media.


This consolidation of clinical, psychological and artificial intelligence knowledge has an immense value to aid clinicians in achieving faster and more accurate diagnoses.