Research On The Construction Method Of Educational Knowledge Graph

With the rapid development of artificial intelligence and internet technology,the education industry is undergoing unprecedented reform and innovation,and a large amount of educational data has been accumulated in various educational information systems.Knowledge graph is a way to describe knowledge with graph structure.By constructing knowledge graph,we can integrate various scattered educational data in the field of education and establishes the correlation between concept entities and educational resources.Educational knowledge graph can bring many intelligent applications to education,thus helping teachers to achieve accurate teaching and students to achieve adaptive learning,which is expected to realize the real ’individualized teaching’.Knowledge graph in education is also known as’concept map’.concept entities in education can be defined as k-grams in educational data(textbooks,teaching aids,teaching video subtitles,test papers,etc.),which should be semantically and syntactically correct phrases that can represent knowledge concepts in related disciplines.Such as:trigonometric functions in mathematics,one-dimensional equations,etc.There are multiple relationships between concept entities simultaneously,while the most common relationships are the sequential learning relationship and dependency relationship between concept entities.The construction of educational knowledge graph mainly involves the extraction of concept entities and the discrimination of relations between entities.How to construct an educational knowledge graph through machine learning technologies such as natural language processing is one of the focuses and difficulties in current intelligent education research.The research content and contribution of this dissertation can be summarized as follows:For the task of concept entity extraction,by combining the characteristics of educational data and using structured information in the domain,this dissertation proposes a concept entity extraction model.This model integrates the global topic information and local clue-word guidance model to efficiently extract concept entities based on the attention mechanism.In addition,in order to make better use of the local clue word information,a soft matching module is pre-trained to improve the generalization ability of clue words because there are many kinds of clue words,and it is difficult to collect complete clue words and clue words do not have generalization.Especially in the case of a small amount of labeled data,the effect of extraction is much better than the baseline model,which greatly reduces the dependence on the amount of annotated data.As for the problem of extraction of inter-entity relations,this dissertation focuses on two types of relations between concept entities:sequential learning relation and dependency relation.On the basis of concept entity recognition,this dissertation proposes the basic features for discriminating the relationship between concept entities combining external knowledge base(Baidu Encyclopedia),and then utilizes machine learning algorithms to achieve efficient discrimination between entities.The relevant experimental results on real datasets demonstrate the effectiveness of the constructed features and the reliability of the classification model.