Exploratory analysis of a large dataset of educational videos: Preliminary results using people tracking

Abstract

The impact of data processing and analytics in education has seen a significant increase in the last decade. The volumes of data produced in education are constantly on the rise, not only in online education, but also in more formal settings. Learning analytics has concentrated mainly on mining patterns in student and teaching interactions by using data logs and texts (e.g., discussion forums, chats, social networks) which are widely available in online learning platforms. However, videos from educational settings have received a smaller interest from the learning analytics community. Educational videos are becoming popular, and there are large volumes shared on different social networks and video platforms. At the same time, deep learning provides powerful techniques for understanding the content in educational videos. In this paper, we propose an exploratory data analysis using educational videos from the YouTube-8 M dataset, one of the largest video datasets and most varied to date. Our preliminary study uses state-of-the-art people tracking neural models to extract features that are then used to cluster educational videos based on the number of people involved. This allows us to identify various educational activities in the YouTube-8 M dataset and to estimate the distribution of these activities from videos uploaded by teachers and learners worldwide.