Research On Campus Online Behavior-aware Student Performance Prediction

With the continuous promotion of campus information construction,colleges and universities at home and abroad have gradually accumulated a large amount of students’campus behavior data.How to analyze students’ behavior habits through these data,predict their performance,and provide suggestions and guidance for school educators and managers has become the focus of attention in colleges and universities.The purpose of performance prediction is to use the historical behavior data in students’ study and life to predict their future academic performance.The student performance prediction method can predict the student achievement in advance and provide the basis for individualized teaching and individualized teaching.At the same time,the school knows the students who may fail in advance and intervenes in advance to help them pass the exam,so as to pursue a better future.Therefore,whether the student achievement prediction method is effective and accurate is of great significance to colleges and universities.According to the theory of psychological pedagogy,students’ learning ability can be judged by continuously paying attention to their behavior habits.As an important part of campus life,network education is rich in resources and is a favorable means for everyone to learn and communicate.With the improvement of campus network authentication system,students’ online behavior on campus is recorded in a hidden way.Taking the perception of campus online behavior as the starting point,this thesis collects and processes students’ online behavior data,and excavates students’ online behavior logs to predict their grades.Therefore,this thesis collects and constructs a real data set containing students’ online behavior and academic performance data on campus,processes the data,and proves that there is a certain correlation between online behavior and academic performance through a large number of data analysis.On this basis,this thesis proposes an end-to-end dual-level self-attention network(DEAN),which introduces a cascade self-attention mechanism to extract students’ local online behavior characteristics every day and long-term global online behavior characteristics respectively,so as to better solve the problem of modeling long online behavior on campus sequence.Through extensive comparative experiments and ablation experiments on the real data set proposed in this paper,the experiments verify the effectiveness of the proposed DEAN network compared with the traditional sequence modeling method.In addition,through data analysis and experiments,it is found that there are characteristic differences between different students,and the same online behavior has different effects on different students.Therefore,this thesis improves DEAN network,encodes students’ online behavior,gives different weights to different behaviors,and pays attention to the characteristic differences.At the same time,multi task learning strategy is introduced to solve the problem of student performance prediction for different majors under a unified framework,and a cost sensitive loss function based on poor student ranking is designed to further improve the performance of the method.Through a large number of experiments,it is verified that the improved DEAN network has better prediction accuracy.