Multimodel data processing method for solving multimodal video lecture processing task

作者：M. E. Ismagulov · 发表于：Highly available systems · 年份：2026 · DOI：10.18127/j20729472-202601-17

Problem statement. Multimodal video lecture–to–text transformation is commonly addressed using multimodal large language models; however, their application is limited by context loss in long videos, low interpretability of the results, and high computational and data requirements. Objective. To develop a multimodel pipeline-based method for multimodal video lecture transformation, in which a video lecture is decomposed into separate modalities processed by specialized models, while a large language model is used only at the final stage for text structuring. Results. A multimodel transformation algorithm has been implemented in the form of a multi-agent system, providing a complete processing cycle for one of the target video lecture formats. Practical significance. The proposed method can be applied to automatic generation of structured textual representations of online courses, preparation of educational materials, and automated documentation of webinars and scientific events.