Construction And Application Of A Knowledge Graph For Educational Assessment

Knowledge graph aims at representing the real entities in the objective world and depicting the relationship between them.Since Google proposed "Google Knowledge Graph" in 2012,knowledge graph has attracted wide attention in academia and industry.Educational assessment aims at describing students' development in all aspects,classroom teaching and learning in a quantitative way,and making value analysis and value judgment on students' behavior,development trajectory and other factors from a qualitative level after collecting certain quantitative data.It is an important tool for implementing scientific and intelligent educational management and research.With the maturity of knowledge graph technology and the intelligent development of educational assessment,people gradually turn their attention to the organic combination of the two.In view of the lack of systematic organization and management of information in the field of educational assessment and the fact that existing assessment often only involves specific numerical statistical calculation,this paper carried out research in the following two aspects:(1)In order to support the interconnection and sharing of information resources in the field of educational assessment,this paper proposes and constructs the Educational Assessment Knowledge Graph(EAKG)covering elements such as schools,students,examination papers,questions,knowledge points and various assessment indicators,and gives the overall structure framework,construction method and process of EAKG.The construction methods of EAKG can be divided into the construction of EAKG schema layer ontology and the construction of specific instances of EAKG data layer.Among them,the EAKG schema layer ontology explicitly formalizes the concepts,attributes,relations,axioms,etc.in the field of educational assessment,mainly including the definition of concept class classification structure,attribute definition and the definition of multiple relations.The EAKG data layer,which is based on the EAKG schema layer structure definition,describes the correlation between objectively existing entities in the field,including the concrete instance of the concept class and the generation of the attribute value of the instance on each attribute.The schema layer-data layer construction method makes the EAKG construction hierarchical and extensible,that is,the new evaluation data can be added to the existing knowledge map as long as it is defined according to theschema layer structure.Furthermore,logical reasoning was performed on the constructed ontology EAKG-withschema,explicit expression of the implied knowledge contained in the declaration was made,and a series of operations such as manual inspection and verification were performed on the inference results.Finally,an EAKG-inferred knowledge graph of educational assessment with 419010 nodes and 3838949 edges was obtained.(2)For the discretization-based EAKG's inherent discretization characteristics and the actual educational assessment often involve the need of numerical statistical calculation,we combine the current mainstream knowledge graph representation learning model to represent the knowledge of logical symbols in EAKG.The knowledge is embedded into the continuous dense low-dimensional vector space and the distributed representation of the entity and relationship is obtained,thus realizing the computational knowledge application in continuous numerical space,such as student entity similarity calculation,test score prediction,knowledge point score prediction,student cluster analysis and so no.Specifically,based on the current mainstream six representation learning algorithms,EAKG(EAKG-noSchema)with only the data layer and EAKG-inferred are represented and compared with their performance on the triple classification and link prediction tasks.The experimental results show that the six representation learning models on EAKG-inferred perform better than EAKG-noSchema on both tasks,indicating the effectiveness of the EAKG construction method of the proposed schema-data-inference layer.Through case analysis,we further compare the characteristics of EAKG's symbolic representation and distributed representation,and compare the EAKG with the existing mainstream large-scale knowledge graphs OpenKN,Freebase and Wiki with the recently proposed big knowledge 10 MC model.EAKG as a whole conforms to the definition of 10 MC model for big knowledge.Based on the background of educational assessment and the popular knowledge graph technology,this paper systematically modeled the key elements in the field of educational assessment,such as schools,students,examination papers,questions,knowledge points and various evaluation indicators.A knowledge graph of educational assessment was proposed and constructed,and it was expressed and learned by using the current mainstream representation learning algorithm.In order to promote the development and application of intelligent educational assessment,various kinds of knowledge applications in numerical space are analyzed.