DBT-wellMind-GameTree: a game-theoretic framework for student mental health support using LLMs

Nowadays, student mental health problems are rapidly increasing in our society, which negatively affects academic success and student self-development. Despite the promise of AI-powered chatbots, existing systems have not adequately explored integrating Dialectical Behavior Therapy (DBT) with game-theoretic modeling to support mental well-being and multi-step decision-making in dynamic DBT path adaptation for intervention. This study proposed the DBT-wellMind-GameTree framework to address this gap. This framework is developed using LLMs, Game theory with game-tree-based mental states, and Dialectical Behavior Therapy. In this DBT-wellMind-GameTree framework, we integrate all DBT components and multi-step planning, enabling dynamic adaptation of the DBT path during the intervention. This dynamic adaptation of multiple mental states uses a game tree to measure the effectiveness of each DBT component. During gameplay, after receiving the user’s utterance and context, the proposed system aims to understand and generate multiple mental states for each DBT component. Then again, generate a reward and probability score that reflects the mental-state-addressing capability of each DBT component for utility-based selection of the optimal DBT component for response generation. The research findings indicate that the proposed DBT-wellMind-GameTree with gpt-4o obtained a mean score for the goal achievement, satisfaction, and coping strategies metric (5.16, 5.03, and 5.19, respectively). In addition, the proposed DBT-wellMind-GameTree with Claude 3.7, this study found mean scores of 5.65 for the goal achievement metric, 5.58 for the satisfaction metric, and 5.61 for the coping strategies metric. Furthermore, this study found sufficient evidence that the proposed framework was statistically more significant than baseline 1 and 2, and outperformed other state-of-the-art methods. The results indicate that the proposed DBT-wellMind-GameTree approach is effective for enhancing student well-being by encouraging acceptance of the situation, validating, promoting effective coping strategies, and enabling better emotional management. Despite a few limitations, this conversation system delivers substantial multilingual support, empowering students and offering valuable mental health resources for effective intervention.