AI-ENHANCED EDUCATIONAL SHORT VIDEOS: KNOWLEDGE DISSEMINATION AND LEARNING EFFICIENCY

Abstract

Current educational short videos generally suffer from problems such as loosely structured knowledge presentation, excessive learner workload, and low knowledge retention. To address these issues, this study proposes an Artificial Intelligence (AI) enhanced educational short-video framework based on multimodal analysis and dynamic AI adaptation. This framework utilizes pre-trained VideoMAE, Whisper, and fine-tuned Contrastive Language–Image Pre-training (CLIP) models to parse and align video content across visual, audio, and text modalities, generating fine-grained semantic indexes. Furthermore, a subject-specific knowledge graph is constructed based on structured segments, enabling adjacency relationships and cross-module association recommendations. A Long Short-Term Memory (LSTM) network is designed for real-time diagnosis of learners' attention states and knowledge mastery levels. Finally, a rule-driven dynamic content adaptation engine triggers personalized intervention strategies based on diagnostic results and the graph topology. This method significantly improves learning outcomes. In the reported learning experiment, the experimental group demonstrated greater improvement in post-test scores, with a 31.6% reduction in cognitive load, a 40.2% increase in fixation time in important domains, and high knowledge retention. This effectively promotes long-term memory and the comprehensive application of knowledge.