An emotion annotated dataset of code-mixed mental health counseling and relevant baselines

Abstract

In India, the realm of mental health is overshadowed by stigma, lack of awareness, and outdated methods, worsened by a scarcity of accessible well-being and counseling solutions. This challenge is particularly prominent in a linguistically diverse nation with over 780 languages, where prevailing research predominantly concentrates on English, disregarding the requirements of the multilingual society. To tackle this issue, our research introduces IndieMH, a distinctive dataset of mental health counseling dialogues embedded in code-mixing, customized to accommodate India’s linguistic diversity. IndieMH, featuring emotionally annotated code-mixed conversations, stands as a valuable asset for both natural language processing and mental health investigations. We conducted thorough validations to ensure the quality of the data, offering insights into the dataset’s potential contributions to advancing mental health support systems and comprehending code-mixed languages in India. Our dataset is compared against 11 benchmarks for emotion classification, laying the groundwork for the development of culturally sensitive AI-driven mental health interventions.