Darya Hryhoryeva

Mental health challenges are widespread, this causes many individuals to remain undiagnosed or untreated due to limited access to care. Advances in NLP, especially large language models, offer new opportunities to identify early signs of distress and to provide scalable and potentially low-cost support through systems such as chatbots and virtual companions. This PhD project will focus on developing and evaluating NLP models capable of detecting linguistic markers of mental health states while maintaining transparency and fairness. A particular emphasis will be placed on understanding model behavior -- explaining why certain predictions are made, how they relate to psychological constructs, and whether they differ across demographic groups. The project will also investigate methods for ethical evaluation, ensuring that predictive or generative models avoid harm and respect user privacy. In the longer term, the research aims to bridge the gap between computational and clinical communities by creating methodologies that are not only technically robust but also clinically meaningful and interpretable for practitioners. The expected outcome is a framework for building trustworthy NLP-based tools that can assist early detection, therapeutic feedback, and responsible deployment in real-world mental health applications.