Predicting end-of-session actions considering the information of learning materials

Abstract

To provide learners with a better learning experience in online educational systems, it is meaningful to understand and model learners’ actions. The actions of learners ending their learning sessions and leaving the systems, which we denote as the end-of-session actions, is important to understand. Modeling the end-of-session actions can lead to useful applications, such as optimizing the way learning materials are presented and interventions that can appropriately help learners. This paper addresses the problem of predicting end-of-session actions in online educational systems. While previous studies have mainly focused on the learners’ behavior in the systems, this paper focuses on incorporating the information of learning materials into the prediction model. Learning material features were extracted by considering multiple perspectives in the learning materials, including their order in the course and their texts. The experiment was conducted using actual user log data from the programming learning system. The experiment demonstrates the effectiveness of incorporating learning material features into the prediction models and analyzed their contribution to the prediction accuracy.