Research And Implementation Of Learning Behavior Analysis And Academic Early Warning Technology Based On Campus Network Data

The development of science and technology has promoted the process of campus informatization and brought a huge amount of student data.With the help of data mining technology in the education field,we can mine useful information for educational managers and learners from massive data,furthermore,to discover and solve various problems in educational research and practice.Using student learning behavior data to evaluate student academic status and curriculum risks,and sending timely academic warnings,has become an urgent need to ensure the quality of college education and reduce the dropout rate.However,most of the existing academic analysis and early warning studies are based on online course only,which lack of the impact of students' behavior habits,making their academic and curriculum risk predictions limited.In response to the above problems,first of all,for the students'academic early-warning problem,this thesis uses the campus network log data to analyze the student's learning behavior,and proposes a Multi-Semantic Feature based Time Convolution Capsule Network(MSF-TCCN)academic risk prediction algorithm.Based on the analysis of data distribution and feature correlation,this thesis designed a multi-semantic layer feature input,and use the vector neuron structure to capture the interaction of student behaviors in time dimension.Experimental results show that the performance of this algorithm is better than many baseline algorithms.Secondly,to address the issue of risk early warning for specific courses,this thesis proposes a Multi-Dimensional Time Feature based Fusion Neural Network(MDTF-FNN)algorithm,which designed a multi-dimensional feature-based time series to input the online learning platform's learning behavior data.This multi-dimensional feature-based time series input is then fused with the daily behavioral characteristics in the network log data to make predictions for the specific course.And the algorithm turns out to be effective through comparative experiments.Finally,a learning behavior portrait and academic early warning system were set up.And the results of this research were demonstrated using Spring framework,Vue.js framework together with the front-end and back-end separation designs.