Knowledge-augmented emotion detection in code-mixed mental health counseling conversations

Abstract

This B.Tech project endeavors to advance mental health counseling through technology by introducing the IndieMH dataset, a curated collection of code-mixed Hindi-English counseling dialogues annotated with emotion labels and translated into Hinglish. We propose a novel framework for emotion classification that integrates ATOMIC and HEAL knowledge graphs to enrich understanding with commonsense and mental health insights. Our exploration encompasses various model architectures, including contextual learning, prompt tuning, and GRU-based approaches, to effectively incorporate extracted knowledge. By leveraging insights from natural language processing and mental health research, our interdisciplinary approach aims to bridge linguistic analysis and mental health insights. This project contributes to the development of emotion-aware technologies in mental health counseling, potentially improving support systems for linguistically diverse communities. It highlights the transformative potential of technology in advancing mental health care and support services.