Estimating Mental Health from Smartphone and Social Media Data: LLM and Machine Learning meets Mental Health

Abstract: Early detection of mental health issues is crucial for preventing their escalation. This seminar presents various approaches to assess mental health using data generated from smartphone and social media usage. To understand the intricate relationship between individuals' mental health and behavior on social media or smartphones, numerous studies employ machine learning, deep learning, and Large Language Models (LLMs) for mental health estimation. Initially, we explain smartphone-based mental health estimation via machine learning, showcasing our research that gauges stress levels from smartphone usage patterns, utilizing the biometric index (LF/HF) as a stress indicator. Subsequently, we explore social media-based mental health estimation through machine learning and LLMs. Specifically, we discuss our involvement in the CLPsych2024 shared task, which focuses on identifying evidence of suicide risk from online mental health content. Introducing a method that merges a supervised approach with BERT finetuning, our team emerged as the top performer out of 17 teams in the highlight extraction task.