Article: Value-oriented meta-adaptive reinforcement learning for optimising emotional intervention Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.50 pp.35 - 51 Abstract: Emotional intervention plays a crucial role in mental health support, yet traditional approaches often lack the dynamic adaptability to individual states and contextual changes. To address these limitations, this study proposes a value-guided meta-adaptive reinforcement learning framework. By integrating meta-learning with deep reinforcement learning, this approach enables intervention strategies to rapidly adapt to users' real-time emotional states and long-term needs. We design an attention-based meta-policy network to extract shared representations across users and introduce a value function to quantify long-term psychological benefits. Furthermore, the framework employs proximal policy optimisation for policy training and dynamically adjusts hyperparameters through a meta-adaptive mechanism to handle non-stationary intervention environments. Experiments on simulated and real-world user datasets demonstrate that the proposed method achieves approximately 22% higher emotional improvement rates and 33% faster convergence speed compared to the best baseline. Inderscience Publishers - linking academia, business and industry through research

Title: Value-oriented meta-adaptive reinforcement learning for optimising emotional intervention

Authors: Bing Lin

Addresses: Faculty of Teacher Education, Zhangzhou City Vocational College, Zhangzhou, 363000, China

Abstract: Emotional intervention plays a crucial role in mental health support, yet traditional approaches often lack the dynamic adaptability to individual states and contextual changes. To address these limitations, this study proposes a value-guided meta-adaptive reinforcement learning framework. By integrating meta-learning with deep reinforcement learning, this approach enables intervention strategies to rapidly adapt to users' real-time emotional states and long-term needs. We design an attention-based meta-policy network to extract shared representations across users and introduce a value function to quantify long-term psychological benefits. Furthermore, the framework employs proximal policy optimisation for policy training and dynamically adjusts hyperparameters through a meta-adaptive mechanism to handle non-stationary intervention environments. Experiments on simulated and real-world user datasets demonstrate that the proposed method achieves approximately 22% higher emotional improvement rates and 33% faster convergence speed compared to the best baseline.

Keywords: meta-adaptive reinforcement learning; affective computing; personalised intervention; proximal policy optimisation; PPO.

DOI: 10.1504/IJICT.2025.151072

International Journal of Information and Communication Technology, 2025 Vol.26 No.50, pp.35 - 51

Received: 28 Sep 2025
Accepted: 30 Oct 2025
Published online: 12 Jan 2026 *

Title: Value-oriented meta-adaptive reinforcement learning for optimising emotional intervention

Keep up-to-date