Cross-modal adaptive reconstruction of open education resources

Matching vast online resources to individual learners' needs remains a major challenge, especially for adults with diverse backgrounds. To address this challenge, we proposed a Dynamic Knowledge Graph-enhanced Cross-Modal Recommendation model (DKG-CMR) to solve the problem. This model utilizes a dynamic knowledge graph-a structure organizing information and relationships-that continuously updates based on learner actions and course objectives. DKG-CMR focuses on three key improvements: (1) Aligning meaning across different data types (e.g., text, video, user behavior logs). (2) Maintaining the knowledge graph's real-time relevance. (3) Reducing the cognitive demand of recommendations (optimizing cognitive efficiency). Our approach employs contrastive learning (a technique for similarity learning) with an enhanced algorithm. It achieved high accuracy (F1-score = 0.912) in multimodal understanding, significantly outperforming baselines (+ 33.7%). The dynamic knowledge graph improved recommendation accuracy by 35.5% while achieving low system latency (1.45 s average, 99% of responses ≤ 1.8 s). Evaluation with 1,520 adult learners demonstrated significant improvements: Participants reported a 40.5% reduction in perceived mental workload (measured by NASA-TLX, p < 0.001). Resource screening time decreased by 56.8%. Mediation analysis identified reduced cognitive load as a primary mediating factor, explaining 47.6% of the total effect variance. We established a Cognitive-Friendly Recommendation (CFR) criterion balancing accuracy with operational efficiency. Implemented in an electronics course restructuring, this work provides an effective framework for techno-cognitive collaborative optimization. Integrating cognitive science insights with cross-modal AI demonstrates significant potential for enhancing resource accessibility and personalization in open education.