The method of personalized foreign language teaching resource recommendation based on MTL-DQN

Abstract

Dynamic and accurate resource recommendation remains a critical challenge in personalized foreign language teaching. Traditional algorithms struggle with dynamic behavior adaptation and long-tail resource neglect. An integrated framework combining multi-task learning and a deep Q-network addressed these challenges. The multi-task learning module extracted shared latent features from structured interaction logs to formulate user learning states. Concurrently, the deep Q-network module optimized recommendation strategies dynamically through continuous environment exploration and delayed user feedback. Experiments evaluated the framework on the EdNet dataset against traditional algorithms including collaborative filtering and graph neural networks, as well as advanced models. The integrated framework achieved a coverage rate of 95.0 percent, an AUC of 0.95, and an F1 score of 0.91. Computational analysis verified an inference speed of under 15 milliseconds per query. The synergy of shared multi-task representations and dynamic reinforcement learning effectively overcomes recommendation latency and data sparsity, and provides a highly accurate, deployable solution for complex educational platforms.