This WebApp has been developed as part of my recent collaboration with a group of psychologists from the Department of Clinical Psychology at University of Rajshahi.
Developed some deep learning models from the ground up based on the Recurrent Neural Network (RNN) model with Long Short-Term Memory (LTSM) and trained on the labeled text to detect eight cognitive distortions. The accuracy values of the models were 0.76 for overgeneralization, 0.85 for all or none thinking, 0.80 for mind reading, 0.83 for fortune telling, 0.80 for labeling, 0.93 for should statement, 0.86 for emotional reasoning, 0.97 for personalization. A web app powered by these deep learning models is able to identify and classify eight types of cognitive distortions. The web app provides a user-friendly interface for inputting Bengali text and receiving predictions about the presence of cognitive distortions. This tool has the potential to be used by therapists and counselors to aid in the assessment of cognitive distortions in Bengali-speaking clients. To our knowledge, this is the first work in Bengali language to detect cognitive distortions using AI.
For training models to detect different cognitive distortions, a relatively large dataset of 4400 comments was constructed. Three dedicated annotators collected them from comments in online news portals on Facebook from the posts from 1 January 2023 to 31 June 2023. They collected data from four of Bangladesh's most commented Bangla news portals.
