Predicting Learner’s Performance Through Video Viewing Behavior Analysis Using Graph Convolutional Networks

摘要

Using videos as a learning resource has gained a great deal of attention, and turn widely used as an effective learning tool. Although predicting the performance of learners seems to be more challenging because of large amount of data in the educational database. In this context, we study the influence of video viewing behavior in relation with learners’ performance, in order to predict whether or not a learner will succeed pedagogical video courses. Indeed, we’re not concentrating on the type of clicks that learners made, but we’re focusing on the pedagogical sequences in which those clicks were made. For this, we have established an experience that begins with selecting educational videos, to which we apply a specific content segmentation into pedagogical sequences. Thereafter, we collected learners’ clicks and their final grades to fill in our database. To analyze this data in order to predict learners’ performance, we used text graph convolutional networks (Text GCN). The Text GCN results achieve an average accuracy of 67.23%. These results show that our approach can make an acceptable prediction of learners’ performance.