Analysis And Application Of College Student Behavior

The relevant policies of education informatization in China clearly put forward the application of big data in college teaching and management.It is feasible because of the vigorous development of big data technology and the massive data accumulated by using information systems in colleges and universities over years.Nowadays,mining the general rules from the massive educational data by big data technology and further promoting the reform of education and teaching,is an important and hot research topic.The research in this article focuses on the four aspects of students' academic,life,mental health and career choices,base on domestic and foreign research,the status of educational data and the actual needs of education and teaching management.This article systematically studies one key issue on each of them,which are predicting academic performance,mining family economic status,the impact of mental health status on social behavior,and predicting career choice.This dissertation will be divided into four parts to study the above four issues.The first part of this article studies the behavior patterns of college students.Using data generated by college students using the information system provided by the school while in school,based on the big five personality trait theory proposed by psychology,diligence,politeness,sleep patterns,consumption behavior features,behavior law features,etc.are extracted.whice have universal meaning.Then the social behavior of college students is modeled based on co-occurrence frequency,and the offline social network of students is constructed.The features of economic level of this network is analysed.Further a social network sentiment analysis model is constructed.This research is a basic research work of the research problem in this article.The second part of this article studies predicting academic performance.The performance prediction in this article is to predict the ranking of students' performance.We abstract this problem as a sorting problem.It is found that students' diligence,orderliness,and sleep pattern have a significant correlation with academic performance by studying the correlation between student behavior data and academic data.Based on results of this analysis,a multi-task ranking learning model(MTLTR-APP)is designed to predict student academic performance,considering the correlation between majors,the dependence on time between semesters and the similarity of students' behavior.The model is advanced in predicting students' performance by training on one grade and testing on the next one in a certain college.The results of experiments show the importance of the correlation between semesters,the correlation between majors and the similarity of students' behavior on this issue.Base on the results,educators can provide early intervention and guidance to students with poor grades or abnormal changes.The third part of this article studies behavior-based anomaly detection of college students.It mainly includes the detection of poverty abnormality based on neural network and the detection of depression abnormality based on propagation dynamics.Poverty anomaly detection is achieved by dividing students into different poverty levels,and we abstract it as a classification problem.Based on the research in the first part of this article,the C4.5 algorithm is used for feature selection.According to temporal characteristics of student behavior data,an algorithm called CW-LSTM with advantages of CW-RNN and LSTM is proposed to establish the model for mining college students' family economic status.Furthermore,the results of experiments on actual students' data of a certain college show effectiveness and advancement of this model.Depression anomaly detection is achieved by analyzing the correlation between students' social behavior and mental health status,and we abstract it as a communication problem.The offline social network of students is constructed using the research content of the first part of this article.It is found that students without obvious depressive symptoms are better at socializing based on the test data of “SCL-90 Assessment Scale”,through analyzing the impact of depression on the structure of the network.Finally,heterogeneous mean field theory is used to describe the model of information propagation.Through computer simulation experiments,it studies the impact of students' mental health on the process of information propagation in college.The results of experiments show that students without obvious depressive symptoms are more likely to receive information.The fourth part of this article studies the prediction of career choice based on integrated learning.Career choice prediction is to predict the four choices of students after graduation.Similar to poverty anomaly detection,we abstract it as a multi-classification problem.Combining with the first research content of this article,it is proposed that factors affecting career choices of college students are professional skills,behavioral rules,interest preference and family economic status.On this basis,two different frameworks are proposed to predict the career choices of college students.The first one is a supervised career choice prediction framework based on Adaboost,whose effectiveness is proved on actual student data of a certain college.And it is found that the mastery level of professional skills,regularity of behavior and family economic stauts are significantly related to career choices.The second framework simplifies the extraction of four types of these features on the basis of the first one,and adds the group divergence of students by generating virtual cluster centers,and then predicts career choice of college students based on the improved XGBoost algorithm(ACCBOX).This framework adopts the method of generating virtual cluster centers to expand the training set,and meanwhile introduces regular association to handle the divergence between real data and virtual cluster centers.Approximate results compared to the first one is obtained on a real data set.Base on the results,educators can optimize student career planning,and consultants martering more information can provide personalized career counseling for students.Also,students who may be at risk of unemployment could be detected,and educators should provide them with targeted employment assistance.