A Tool for the Semantic Analysis and Recommendation of videos in e-learning

Abstract

Abstract-Video lessons are increasingly adopted in education, especially in universities and lifelong learning projects. Their popularity is due to the people's familiarity with video and to other intrinsic characteristics of this medium, such as the message rapidity and its reproducibility. Accordingly, Massive Open Online Courses are gaining a prominent role in both formal and informal education and many universities provide video courses for their students through suited platforms or even freely accessible to everyone. To improve the effectiveness of video lessons and to make them part of a wider learning environment, we decided to investigate the possibility of making a system that, starting from a video, can suggest further "readings". Such a system is thought for independent lifelong learners, for regular students, for teachers and instructional designers as well.

A Closer Look into Recent Video-based Learning Research: A Comprehensive Review of Video Characteristics, Tools, Technologies, and Learning Effectiveness

“…A popular approach was to extract topics, concepts, and keywords to compute the similarity to other videos. The use of ontologies and external knowledge bases was used particularly to extract concepts [24,25,40,206]. Some interesting approaches considered other characteristics such as video readability [111],…”

“…Regarding tasks and technologies, we identified some low-level tasks performed as basic steps, of which, many rely on deep learning approaches. For example, for textual video elements, several studies employed optical character recognition [14,21,29,32,49,50,54,64,67,102,109,114,123,138,152,160,161,179,195,255,261,262,265,273,276], keyword extraction [14,40,43,61,92,105,106,109,112,121,128,131,161,206,255,264,271], generic natural language processing methods (e.g., [29,127,128,194,243]), or utilized word embeddings (e.g., [54,…”

“…For image-related data, many studies suggested approaches for slide transition detection and video segmentation [17,29,32,64,67,69,71,72,77,83,110,115,153,160,178,205,220,262,265,273]. For audio tracks, various studies carried out speech recognition to generate transcripts [14,16,21,24,29,31,40,54,64,69,71,102,106,109,142,161,182,202,250,267,273]. In the next subsections, we discuss the tasks and technologies in detail according to each type of tool.…”

“…and Vercelli [40] extracted concepts and also related concepts by linking the content to Wikipedia articles. The learners could receive recommendations of videos and also Web pages.…”