Research On Performance Prediction Based On Knowledge Graph

In education-related research,students’ academic performance is often seen as one of the important indicators of their learning outcomes and overall quality.By studying students’ learning performance,we can not only intervene in advance of students’ learning situation,but also provide scientific basis for the setting of training plans.Taking this as the starting point,this paper integrates the knowledge graph into the process of performance prediction,uses the rich semantic relationship of the knowledge graph to explore the semantic relationship between different courses,and sorts out and refines the sequential relationship between courses through entity relationship extraction,to improve the accuracy of grade prediction and the interpretability of prediction results.This paper mainly does two aspects of work: complete the construction of the knowledge graph of computer science courses;A student achievement prediction method based on course knowledge graph is proposed and its effectiveness is verified.First,according to the undergraduate training plan of the computer school of a university,the course introduction and learning guide text of the corresponding national excellent courses are crawled from the MOOC website.The unstructured course introduction and learning guide text obtained are preprocessed by using Chinese stop word list and self-constructed computer professional vocabulary,calculating the keyword weight by TF-IDF(Term Frequency–Inverse Document Frequency)algorithm,and outputting the knowledge points of the top 20 weights to complete the extraction of course knowledge points.The relationship between reference books and courses is extracted by combining machine learning methods and rules,and the triad information composed of the above knowledge points,reference books and course priority relationships is used to construct the knowledge graph of computer major courses based on the Neo4 j graph database.Secondly,based on the preceding course knowledge graph,this paper proposes an achievement prediction model based on the knowledge graph(AP-KG),which employs the knowledge representation model to embed the course knowledge graph in low dimensions and computes the similarity of the embedding vectors via the distance model to obtain the similarity matrix between courses.The semantic similarity of courses calculated based on knowledge graph is fused with the similarity calculated by collaborative filtering recommendation algorithm,and student performance prediction is completed based on the fusion similarity,and the collaborative filtering recommendation algorithm is improved from the dimension of semantic relationship without taking advantage of the defects of semantic association of recommended objects,which can ease data sparsity and cold start issues to some degree while increasing prediction accuracy and interpretability.Finally,through several groups of comparative experiments,AP-KG performs better than other nine classical recommendation algorithms in the data set of student ’scores,and the root mean square error(RMSE)and mean absolute error(MAE)are considerably decreased,demonstrating the AP-KG score prediction method’s accuracy and efficacy,increases the accuracy and explanatory power of forecast findings,and has some implications for encouraging the building of smart campuses and assuring the quality of staff training in colleges and institutions.