Proficiency Prediction System for Online Learning Based on Recurrent Neural Networks
You-Xuan Huang, Nen‐Fu Huang, Jian-Wei Tzeng, James Liang, Ching-Wei Su, Yao-Ting Li · 2022
In recent years, online learning systems have become increasingly popular among students and teachers because they are unlimited by time and space. A brand-new online learning system called QSticker, which is based on a line bot, has also been proposed to improve the online learning environment. However, there is still a problem in the online learning system that teachers and students cannot communicate face-to-face in time and care about students' learning status, so if there is not a proper analysis of students' performance, it will lead to poor learning conditions. There have been many pieces of research about knowledge tracing in the past. Nonetheless, we found that the knowledge tracing models cannot optimally predict students' proficiency in knowledge concepts in some online learning environments such as QSticker. Therefore, we proposed a Proficiency Prediction System based on Gated Recurrent Unit (GRU). Since many students have similar trajectories in learning, the system uses the most straightforward exercise answering behavior data, including the correctness of his answer and the knowledge concept correlations. It then calculates other characteristics, such as the difficulty of each knowledge concept, to predict students' proficiency in all knowledge concepts in the course. The accomplished experiments show that our model can achieve 71% accuracy on the collected dataset. With the help of this system, we can predict the difficulties students may encounter in the learning process. In addition, to be practically used in teaching scenarios, we also designed an analysis platform for this system.