Probabilistic Modeling Towards Understanding the Power Law Distribution of Video Viewing Behavior in Large-Scale e-Learning

In the era of internet, e-Learning has become vastly widespread and generated huge amount of log data of video viewing behavior. Through analyzing and mining these log data, significant Power Law Distribution (PLD) of viewing behavior is observed, which is different from small-scale e-Learning or traditional classroom environment. In this paper, we apply the mechanisms for generating the PLDs in analyzing log data of a large-scale e-Learning platform to discover the factors influencing the video viewing behavior. Firstly, four factors correlated to the video viewing behavior are discovered from log data, including the number of videos viewed, the start date of viewing videos, the date of final exam, and the duration of enrollment. Furthermore, we present a probabilistic model of viewing behavior based on the four factors. Finally, the accuracy of the model is validated with nine online courses in which each course enrolled more than 1,000 students. In addition, we analyze the application of the proposed model and provide some valuable suggestions for teachers to improve the performance of students.