Prerequisite Relation Learning for Course Concepts Based on Hyperbolic Deep Representation

Abstract

With the rapid development of MOOCs, more and more learners participate in online learning to improve their abilities. However, students from different educational backgrounds have different starting points, requirements and foundation. How to use the hierarchical relationship between course concepts to provide learners with personalized learning courses can help learners overcome difficulties and reduce the dropout rate of online learning. course concept has the characteristics of hierarchy and prerequisite relation. We can integrate the prior knowledge of hierarchy into the feature vector of course concept. Traditional methods based on machine learning or deep learning embed the features of course into Euclidean space, thereby losing the embedded feature information of hierarchical relationship between course concepts. In recent years, the deep learning method based on hyperbolic space has unique advantages in representing hierarchical structure data. Based on this, we propose a hyperbolic deep learning representation based course concept relationship prediction algorithm, which can better retain hierarchical relationship information between concepts. First, we use the word embedding method of hyperbolic space to express the course concepts (feature vector extraction), then we has defined the distance of the two vectors of in hyperbolic space and calculate the distance between course concepts, finally we determine whether there is a prerequisite relationship between course concepts through hyperbolic FFNN and hyperbolic MLR.The experimental results show that the hyperbolic space is better than the Euclidean space. (F1 scores improved 6.8%).