Prerequisite Relation Learning From Educational Data And Ordered Knowledge Graph Construction

People often learns in a “prerequisite” order,which means they study knowledge from simple ones to cutting-edge ones.The wrong order of learning will not only increase the difficulty of learning,but also go astray because of the misunderstanding of the subject concepts.However,such a correct and efficient order is often difficult to find.There should be some domain experts manually determining which to be the first after completing the exploration of the subject.This means that getting a “prerequisite” order for a subject not only requires plenty of expert time,but also can barely help the exploration of academic frontiers.This paper takes the automatic extraction of prerequisite relations from educational materials as a start point.It can improve the performance of prerequisite relations recognition task by marking high-quality phrases,use a designed algorithm to extract the curriculum concepts,and integrating multiple information sources for classification.Here follows four main research tasks.(1)High-quality phrase extraction method based on pre-trained model.Thanks to its well-designed neural network framework,the pre-trained language model can efficiently make use of vast unsupervised corpus on the Internet and absorb the hidden knowledge in them.Therefore,construct a high-quality phrase extraction model from the pre-trained model can absorb general knowledge in human society without additional training corpus.Compared with traditional phrase extraction methods,methods based on pre-trained model have achieved better results on educational data.(2)Domain-specific concepts identification method based on graph propagation algorithm.The concepts in same domain often have a certain degree of relevance and similarity in terms of names,and often co-occur in educational corpus.Therefore,using the graph propagation algorithm on a designed graph structure will identify the domain-specific concepts more efficiently.The experimental results prove the effectiveness of this method.(3)Prerequisite relation recognition based on fusion of multi-source information.While making prerequisite judgments of two concepts,People often make us of some fixed information resources.The recognition effect will be affected if single information resource is used.To solve this problem,a prerequisite relation recognition model that integrates multi-source information is proposed.It can express and interact with different types of information from different resources,which improves the effect of prerequisite relation recognition.Experiments show the effectiveness of this method.(4)Construction of an ordered knowledge graph.This paper integrates the above research results into a display system to simulate the effect of an ordered knowledge graph composed of prerequisite relations.In order to evaluate the effect of the system more objectively,manual and automatic evaluation methods are used.In addition,the problems in the system are also analyzed to provide guidance for future work.