Fusion of Deep Reinforcement Learning and Educational Data Mining for Decision Support in Journalism and Communication

The project-based learning model in journalism and communication faces challenges of sparse multimodal behavior data and delayed teaching interventions, making it difficult to perceive student states and optimize decisions in real-time. This study aims to construct an intelligent decision-support framework integrating educational data mining (EDM) and deep reinforcement learning (DRL) to address these issues. A bidirectional long short-term memory (Bi-LSTM) network models behavioral sequences, while a conditional generative adversarial network (cGAN) with Wasserstein optimization enhances low-activity student data. The extracted and augmented features are then fed into a Double Deep Q-Network (DQN) to generate adaptive teaching intervention strategies. Experimental results from a 26-week study show that the proposed framework improved personalized learning-path matching from 0.42 to 0.68, increased knowledge mastery from 40.46% to 77.13%, and reduced intervention latency from 210.5 min to 144.6 min. The results demonstrate that the fusion of EDM and DRL can achieve efficient and adaptive decision-making, providing a viable approach for intelligent teaching support in journalism and communication education.