Rani, Saima (2024) AI as a Mirror to the Mind: Analysing Mental Health Narratives in the Social Media Landscape. Research Master thesis, Victoria University.

Abstract

The intersection of mental health challenges and social media’s broad reach has become a key public health research focus, particularly due to the profound global impacts of the COVID-19 pandemic on mental well-being. This thesis advances mental health research by developing a comprehensive framework for analysing mental health narratives from over one million Reddit posts during pre-pandemic, pandemic, and post-pandemic period. Our novel data annotation approach, integrated with Natural Language Processing (NLP) techniques, uncovers foundational triggers of mental health issues as expressed through social media discourse. Two primary objectives are central to our study: (1) compiling an extensive corpus from social media posts to serve as a benchmark dataset for analysing mental health-related expressions, and (2) employing NLP-based framework for systematically identifying and analysing underlying factors contributing to mental health conditions. This dual approach not only facilitated the categorisation of mental health triggers but also enabled us to explore Artificial Intelligence (AI) models’ alignment with human judgement, assessing AI’s capability in interpreting complex mental health dialogues. Our mixed-methods approach, combining quantitative analysis for identifying temporal trends in mental health discourse during the pandemic era, with qualitative thematic analysis of selected posts for in-depth exploration, offers a nuanced perspective of the mental health landscape. Key findings demonstrate that machine learning (ML) can approximate human decision-making in mental health assessment, revealing critical gaps that AI must bridge to fully capture the nuanced linguistic and emotional contexts of mental health discussions. The public availability of our dataset, including an annotated subset as detailed in the data availability section, is poised to be a significant resource for training ML models. This work enriches academic discourse around mental health diagnostics and intervention, offering actionable insights for effective public health strategies. By delineating the role of AI and NLP, this thesis highlights their potential in advancing our understanding of mental health dynamics, ultimately promoting a more empathetic and effective approach to public health.

Search Google Scholar

Repository staff login