Conversational AI for cognitive distortion detection

Abstract

Cognitive distortions are dreadful, recurring negative thought patterns. They are a silent yet powerful force behind declining mental health that could pave the way for prolonged emotional distress, often culminating in depression if left unchecked. This thesis attacks this growing concern by introducing a Conversational AI-powered application that detects these distortions while engaging in chat exchanges with students.

The research begins with identifying the ten most common cognitive distortions, such as all-or-nothing thinking, overgeneralization, emotional reasoning, labelling, personalization, mind-reading, fortune-telling, magnification also known as catastrophizing, should statements, and mental filter. The functionality ties an LLM to a retrieval-augmented generation mechanism that incorporates a vector database, ensuring contextually grounded responses through recursive document splitting and history-aware retrievers. Ultimately, it integrates prompt engineering techniques to rationalize messages fraught with cognitive nuances from users.

The resulting system takes the form of an intelligent chatbot assistant that provides a supportive space where University of Oulu students can reflect on their thoughts in order to foster cognitive self-awareness and reduce mental distress through soothing chat engagement.