A team of scientists from the University of Alberta built a new machine learning model that can detect early signs of depression in texts similar to tweets from Twitter.
Nawshad Farruque, a graduate student who designed the machine learning model to detect clues in every communication, said: The findings of the research suggest that we can build critical predictive models that can precisely be used to identify signs of depression in texts. While we are using the model to identify depressive language on Twitter accurately, it can also be applied to written texts from other platforms for detecting depression.
The researchers built the model using samples of writing by people who seem to have depression on online depression forums. This new model was then trained to identify depression through linguistic clues on Twitter.
Farruque added: “This is the first study to demonstrate that depressive language has a particular linguistic representation. We have shown that it can be possible to identify the signs, transfer it, and use it in detecting depressive languages.” He noted that the model could be used for several applications, ranging from detecting depression signs earlier to helping psychiatrists monitor the success rates of the treatment for their patients over time.
“Our algorithm can be integrated with a chatbot that can communicate with seniors and can detect signs of loneliness and depression. Another primary application of this model could be to oversee the messages of high-school and college students to identify whether they are suffering from depression or loneliness,“ Farruque said.
Farruque conducted the study under the guidance of computer scientists Randy Goebel, a research fellow from Alberta Machine Intelligence Institute (AMII), and Osmar Zaïane, who is the director of the AMII.