Quantification and prediction of engagement: Applied to personalized course recommendation to reduce dropout in MOOCs

Abstract

MOOCs (Massive Open Online Courses) offer tens of thousands of courses and attract hundreds of millions of online learners. After years of development, these platforms have accumulated a large number of learning action data. These data imply learners’ engagement with the enrolled courses, reflecting whether learners are willing to spend time on the courses. Meanwhile, the MOOC dropout rate remains chronically high. This paper experimentally finds that using the relationship between engagement and dropout rate can help MOOC platforms develop effective methods to reduce the dropout rate. Firstly, this paper proposes a new Quantified Engagement method named QE by using learning action data and learning duration to quantify learners’ engagement in enrolled courses. Next, an Engagement Neural Network prediction model named ENN is proposed to predict learners’ engagement in unenrolled courses. Then, applying QE, the predicted engagement by ENN, and the aforementioned relationship to personalized course recommendations to learners, ensuring that the recommended courses are likely to be completed by learners as much as possible, thus effectively reducing the dropout rate. Finally, the proposed method is evaluated on two large real-world datasets in XuetangX and KDDCUP. The RMSE and MAE of ENN are 0.1066 and 0.0727 on XuetangX and 0.062411 and 0.039621 on KDDCUP, respectively. Dropout rates were reduced by 46.99% and 10.34%, respectively, when 5% of the courses were recommended. These results demonstrate that the quantification method of engagement is valid, applying predicted engagement to personalized course recommendations and reducing dropout rates is available.