Prerequisite Relation Prediction Algorithm For Course Concepts Based On Hyperbolic Space Representation

With the rapid development of online education,more and more learners participate in online courses to increase their knowledge reserves.However,for students with different backgrounds,it is difficult to find the learning order suitable for them,resulting in a low degree of completion of online courses.Therefore,this paper hopes to clarify the sequential learning relationship between the course concepts through the information of the course concepts in the online education platform,so as to find the appropriate learning path of the course concepts for learners,and then help learners better grasp the course concepts and related knowledge.This paper attempts to quantify the relationship between concepts through vector representation learning of the course concepts,and then deduce the sequential relationship between concepts.Among them,the most important work is the representational study of the curriculum concepts and the definition of the conceptual relations.Most of the existing representation learning methods transform data features from high-dimensional space to low-dimensional space based on Euclidean space in order to obtain deeper abstract representation of data features.However,the curriculum concept of the object studied in this paper has a certain characteristic of hierarchical structure,and the representation vector learning based on Euclidean space may lose the characteristic information of the implicit hierarchical structure relationship between the curriculum concepts.This paper hopes to integrate this hierarchical structure information into the representation learning of the feature vector of the curriculum concepts.In recent years,deep learning methods based on hyperbolic space have unique advantages in representing hierarchical data.Therefore,two methods of representation learning based on hyperbolic space are proposed in this paper.(1)Distance model representation algorithm based on hyperbolic space.The equivalence model based on hyperbolic space defines the distance function in hyperbolic space,in order to explicitly capture the hierarchical structure features among course concepts in embedded space,and obtain a more suitable vector representation of course concepts.Then,we applied the learned vector representation to predict the correlation between the curriculum concepts and analyzed the sensitivity of the vector representation dimension to the algorithm.The experimental results show that the modified method has good performance and effectiveness on the three curriculum concept datasets.(2)Graph convolutional network algorithm based on hyperbolic space.The ability of hyperbolic space to capture hierarchical information is combined with graph convolutional network,which has advantages in processing network graph structure data,and attention mechanism is introduced.The experimental results show that the graph convolutional network based on Euclidean space is more effective than the graph convolutional network based on Euclidean space.To sum up,this paper proposes a representation learning method based on hyperbolic space to predict the curriculum concept relationship,so as to better retain the hierarchical structure relationship information among concepts and improve the accuracy of the curriculum concept relationship prediction to a certain extent.