EDEN: Enhanced Database Expansion in eLearning: A Method for Automated Generation of Academic Videos

Rushil Thareja, Deep Dwivedi, Ritik Garg, Shiva Baghel, Mukesh K. Mohania, Jainendra Shukla

Abstract: Academic video databases, integral to e-learning platforms, necessitate continual updates for freshness and diversity. In addressing this challenge, we introduce EDEN, an end-to-end method for expanding academic video databases. When presented with user-defined topics, EDEN retrieves existing videos or, if unavailable, generates new ones in the same style as the original source database, ensuring seamless integration. The system’s efficacy is evidenced by its addition of 3134 videos across diverse K-12 subjects, significantly enhancing an existing database’s scope. Key performance indicators, including a +6% increase in F1 BERTScore and a +9.7% rise in mean Image Visual Relevance score, demonstrate the superiority of our fine-tuned LLM and stable diffusion models over standard versions. Notably, EDEN achieves remarkable efficiency, generating one second of video content in just 0.77 s using consumer-grade GPUs.