Student Performance Prediction Based on Multi-view Network Embedding

Predicting student performance is a very important but yet challenging task in education. In this paper, we propose a Multi-View Network Embedding (MVNE) method for student performance prediction, which effectively fuses multiple data sources. We first construct three networks to model three different types of data sources correlated with student performance, ranging from class performance data, historical grades, to students’ campus social relationships. Then we use joint network embedding to learn the embedding representation of students and questions based on the proposed separated random walk sampling. Student performance is predicted based on both student and question similarities in the low-dimensional representation. Experimental results on the real-world datasets demonstrate the effectiveness of the proposed method.