When it comes to determining depression, doctors usually ask patients specific questions about mood, mental illness, lifestyle and personal history, and use these answers to make a diagnosis. Researchers at the Massachusetts Institute of Technology have created a model that can detect depression in people without having to ask them specific questions, based instead on their natural conversation and writing style.
According to the project's main researcher, Tuka Alhanaai, “the first clues we get about , that a person is happy, agitated, sad or in a special mood, like depression, precisely from a conversation. To deploy a scalable model for determining depression, you need to minimize the number of restrictions that you apply to data. You need to conduct a routine conversation and allow the model to determine a person's state in the process of natural communication. “
How to identify depression in a person? AI enough talk
Researchers call their model “extra-contextual”, because there are no restrictions on the types of questions that can be asked, or the types of answers that will be heard. Using the method of sequence modeling, scientists fed text and sound models from conversations with people in depression and beyond. As the sequences accumulate, patterns such as the natural use of words like “sad” or “down,” and audio signals that are more monotonous pop up.
“The model sees a sequence of words and determines these patterns as more likely to be found in people with depression or without it, “Alhunai says. “Then, if she sees the same sequences in new people, she can determine whether they have depression.” During the tests, the model demonstrated a 77 percent success in determining depression, thus avoiding all other models that rely heavily on structured questions and answers.
Perhaps one day this model will become a tool for doctors or even the basis for future systems of artificial assistants.
Do you think it will help? Tell us in our chat in Telegram.