Deep Reinforcement Learning-based Personalized English Vocabulary Learning Recommendation System: Integrating Learner Cognitive State Modeling

English vocabulary acquisition is a core bottleneck in second language learning, and personalized recommendation systems have become a key tool to optimize learning efficiency. However, existing systems rarely dynamically model learners’ cognitive states (e.g., vocabulary proficiency, memory decay, and learning load) or adapt to real-time changes in learning processes. To address this gap, this study proposes a personalized English vocabulary learning recommendation system based on deep reinforcement learning (DRL), which integrates a dynamic cognitive state modeling framework. First, we construct a multi-dimensional cognitive state evaluation index system, including vocabulary mastery, memory retention, and learning fatigue, and design a quantitative model to characterize cognitive state evolution using forgetting curve theory. Second, we propose a DRL-based recommendation framework (Cog-DRL) that embeds cognitive state features into the state space, designs action strategies oriented to vocabulary difficulty and review frequency, and optimizes the reward function by balancing immediate learning effects and long-term memory consolidation. Finally, extensive experiments are conducted on two datasets (a public vocabulary learning dataset and a self-collected dataset of 520 learners) to compare Cog-DRL with traditional recommendation methods and vanilla DRL models. Experimental results show that the proposed system outperforms baseline models in terms of recommendation accuracy (NDCG@10 improved by 12.3% − 21.7% ), vocabulary mastery rate (improved by 15.6% on average), and learning efficiency (time cost reduced by 18.2% ). This study provides a new paradigm for integrating cognitive science into intelligent language learning systems, offering theoretical support and practical solutions for personalized vocabulary education.