Towards Explainable Artificial Intelligence for Mental Healthcare via Social Media

Mental health challenges present significant global, social, and economic concerns. Mental health conditions may lead to serious consequences, including self-harm, and suicide. In today’s interconnected society, social media platforms provide valuable insight into individuals’ thoughts and emotions. This dissertation explores the use of platforms for the mental disorder assessment. Although existing AI-based methods can identify mental disorders, they often overlook explainability, which limits their practical adoption.

Initially, we propose AttentionDep, a domain-aware attention model that drives explainable depression severity estimation by fusing contextual and domain knowledge. Our experiments demonstrate that AttentionDep outperforms state-of-the-art baselines by over 5% in graded F1 score across datasets, while providing interpretable insights into its predictions. This work advances the development of trustworthy AI systems for mental health assessment from OSM. Further, we develop RedCoM, a new Reddit dataset for analysing contributors to mental health challenges; and MDCNet an innovative multi-class learning framework. Through extensive evaluations on existing benchmark datasets, MDCNet outperforms state-of-the-art methods, achieving significant improvements across all evaluation metrics. These results highlight MDCNet's ability to enhance the contextual understanding of contributors to mental disorders and stress, allowing for more effective screening methods. Lastly, we develop a theoretical framework for XAI in MH applications. We introduce the MentalXAI model for explainable mental health assessment via textual social media data. This contribution allows us to investigate the factors that influencing the model's decision. The evaluation of MentalXAI model has been conducted based on the common mental health conditions. The model consistently outperforms state-of-the-art explainability baselines, demonstrating robust performance across all metrics.

To the best of our knowledge, this dissertation proposes the first comprehensive explainability framework for mental health that covers disorder severity, contributors, and the key features that influence the model’s decisions which play a vital role in developing responsible data-driven systems for assessing mental disorders.