A recent study published in Frontiers in Psychiatry explores how a chatbot could be developed to identify early symptoms of depression in adolescents.
Detecting depression in adolescents through language and AI: Closer than we think?
To do this, Jarvers et al. (2024) conducted structured interviews with 53 adolescents, both with and without a diagnosis of depression, using the Mini-International Neuropsychiatric Interview for Children and Adolescents (M.I.N.I.–KID). From these interviews, the authors generated a corpus of 4,077 question–answer pairs, which served as the basis for training a classification model using BERT vectors (Bidirectional Encoder Representations from Transformers). The goal was to distinguish between adolescents with and without depression based on their verbal responses.
The researchers also explored generating synthetic responses with ChatGPT to expand the dataset and improve model training.
Can synthetic clinical data help improve Large Language Models?
According to the authors, this possibility represents an important opportunity to investigate. The main findings showed that the BERT model reached up to 97% accuracy in distinguishing interviews with adolescents with depression from those without.
As a linguistically relevant outcome, the analyses revealed differences such as a greater use of the pronoun “I” among adolescents with depression and a lower frequency of prepositions. In addition, the researchers confirmed that synthetic data generated by ChatGPT improved BERT’s performance; however, they also identified limitations, such as ChatGPT’s difficulties in understanding long instructions containing abstract descriptions.
This groundbreaking study offers a promising starting point in the use of linguistic analysis and AI to further support early depression detection among adolescents, a group that often faces barriers to accessing mental health services. The findings also confirm that synthetic data can supplement research when real datasets are limited.
Toward a new way of detecting clinical depression in adolescents
The authors propose advancing toward the development of clinical chatbots capable of conducting semi-structured interviews, detecting depressive symptoms, and guiding users toward appropriate intervention. That said, further research will be necessary to ensure data validity and address ethical and privacy considerations when working with adolescents.
