基于多目标优化的MOOCs课程视频摘要生成框架研究

以MOOCs为代表的在线学习技术快速发展，为学习者提供了前所未有的大规模、随时随地、高质量的在线课程视频。然而，由于课程视频具有线性、非结构化、冗长、包含丰富的语音和视觉信息等特性，学习者很难快速了解和定位其内容。因此，学习者在学习过程中持久性不强、专注度不够、效率不高，这些视频资源往往没有被充分利用。本项目根据MOOCs课程视频数据的特点，提出一个基于多目标优化的视频摘要生成框架，分析课程视频，对原始视频内容进行总结，以直观呈现的方式帮助学习者确定课程内容。首先，通过深度学习模型深入挖掘课程视频包含的语音信息，对视频进行主题分割。然后，构建学习者对于课程视频的注意力模型，度量每个主题内视频帧视觉内容的重要性。最后，采用稀疏表示的策略生成具有表示性、代表性和重要性的视频摘要。本项目旨在丰富课程视频摘要生成的理论与方法，提高课程视频理解的智能化水平，既有科学理论价值又有社会应用前景。

The rapid development of online learning technology, represented by MOOCs, has provided unprecedented large-scale, anytime, anywhere, high-quality online course videos for learners. However, because the course video is linear, unstructured, lengthy, and contains rich audio and visual information, it is difficult for learners to quickly understand its content. Therefore, learners are not persistent, focused and efficient in the learning process, and these video resources are often not fully utilized. According to the characteristics of MOOCs course video, this project puts forward a video summary model based on multi-objective optimization to analyze the course videos. The model summarizes the original video content, and enables learners to quickly understand the main content contained in the video in a visual presentation way. Firstly, the deep learning framework is used to dig deep into the speech information contained in the course video and segment the video into different topics. Then, the learner's attention model for the course video is constructed, and the importance of the video frames within each topic is calculated. Finally, sparse representation strategy is adopted to generate video summaries with representability, typicality and importance . This project aims to enrich the theory and method of course video summary generation and improve the intelligence level of MOOCs course video understanding. It has both scientific theoretical value and social application prospect.