The Applied Research Of Student Achievement Prediction Based On Graph Neural Network

Student performance prediction is one of the most important research topics in the field of Educational data mining,which aims to obtain better teaching efficiency and educational output.There is a certain link relationship between the knowledge information between students and between students and courses in educational data.Therefore,a topology map structure can be constructed through complex link relationships to deeply mine the key attribute feature information in educational data.Because the traditional algorithm can’t directly predict the data of graph structure,it can’t pay attention to the complex link between attribute feature information.To this end,this paper based on online and traditional two aspects of education data and depth learning related theories were respectively carried out research,the research content is helpful to realize the academic early warning and teaching intervention for the underperforming students,and provides a strong guarantee for improving the teaching quality.Specific research findings are as follows:(1)Aiming at the problems of traditional performance prediction methods,such as the importance of attributes to students’ achievement and the low completion rate of students’ online learning,this paper proposes an attention mechanism integrated R-GCN network and GRU network algorithm for student performance prediction.Firstly,the information of students and courses is represented by graph structure.Secondly,the attention mechanism is used to capture the attribute features of the relationship between students,and extract and visualize the important attribute features of students,the method synthesizes the advantages of R-GCN neural network and GRU neural network,not only can it capture the internal relations between nodes,but also can extract the most representative characteristic information of student’s behavior,the feature information is spliced,and the performance of the model is improved by optimizing the objective function,and the students’ grades are classified.The experimental results on the open data set show that the proposed method can accurately predict the failing students,verify the effectiveness of the attention mechanism,and bring a positive impact on the later teaching intervention.(2)in order to make full use of the heterogeneous information between the students and the course nodes in the higher education data,through the interaction of the heterogeneous information,to improve the students’ underperformance,in this paper,a prediction algorithm based on type-aware heteromorphic graph neural network is proposed.Firstly,the heterogeneous graph representation is constructed according to the heterogeneity of the data.Secondly,the student and the course node are transformed into the same vector space for neighborhood aggregation by transformation matrix,the attention of different edge types at different node levels aggregated rich attribute information from the neighbor nodes of the course,and perceived additional feature information of different edge types,at the same node level,the attention is focused on the students who are at risk of failing,and the ability of the model to predict the students who are at risk of failing is enhanced by suppressing the irrelevant attributes and adapting the attention mechanism,the extracted features are optimized as a whole by the objective function.The experimental results show that the proposed algorithm can identify the failing students better than other algorithms,thus giving appropriate teaching intervention to the students with poor performance.The thesis includes 22 figures,8 tables and 70 references.