Intelligent Gamified Therapy Using Dilated Residual Network and Bayesian Inference Learning Automaton Based Reinforcement Learning for Emotional Regulation

Abstract

Emotional well-being is essential for maintaining mental health, yet many individuals struggle with regulating negative emotions due to limited access to effective interventions. This study proposes a gamified digital therapy framework using the strategic board game Go to enhance emotional regulation and provide mental health support. The framework integrates a Dilated ResNeXt (Dil-ResNeXt) network for advanced spatial feature extraction and contextual learning, combined with a Bayesian Inference Learning Automaton (BInfLA)-based Reinforcement Learning model to dynamically adapt policies under uncertainty. Additionally, the Predictive Upper Confidence Bound applied to Trees (PUCT) algorithm is employed to improve decision-making efficiency in gameplay. Experimental evaluation demonstrates that the proposed system achieved a 12.8% increase in winning rate stability and improved technical balance scores by 15% compared with conventional AlphaGo-based models. A psychological assessment with 128 participants reported a 20% improvement in emotional regulation and a 17% reduction in anxiety levels. These findings indicate that the proposed framework significantly enhances user engagement and emotional well-being, providing a robust approach to digital mental health interventions.