Research On Social Causality Inference Method Based On Multidimensional Hawkes Process And Its Application

Social relations play an important role in college students' campus life.Positive social relationships will show in a good way the development of college students' mental health,and negative social relationships will cause negative mental effects on college students and even bring about the risk of depression.Excavating college students' social relations and evaluating them quantitatively has always been the focus and difficulty of education research.With the rapid development of big data,cloud computing and Internet information technology,education informatization has become the inevitable development trend of college education management.In the context of education informatization,a large amount of student behavior trajectory data is recorded in the form of swiping card sign-in,which provides huge opportunities and challenges for studying student social relationships and exploring its internal laws.Existing research uses check-in data,based on the idea of co-occurrence,and uses the relevance of movement trajectories to measure social relationships.However,the experimental results of the correlation cannot reflect the effect of the two behaviors,and often meet explanatory deviations.The coincidental co-occurrences of students due to similar work and rest patterns will make the results unconvincing.In order to solve this problem,this thesis uses the multidimensional Hawkes process to describe the behavior of students' swiping card,explores the Granger causality of college students' social relationships and studies its application.The main research content and innovation work include:(1)A method of social causality inference based on the multidimensional Hawkes process is proposed.This thesis uses the campus card data to conduct modeling based on the multidimensional Hawkes process.Considering the homogeneity of the student population and the heterogeneity of the swiping location selection,preference coefficients are set.Combining the sparsity and low rank of the social network as a penalty item,it explores the social relationships hidden in the student's campus swiping behavior.In order to verify the validity of the model,using the card data of 65 college students from two classes in the school of Physics of a key university,the results estimated the Granger causality diagram between college students.Comparing the data from the student social relationship questionnaire survey,the model's inference accuracy and recall rate reached 98%and 89%,respectively,showing good performance.Finally,this thesis uses the exponential random graph model and the separable time exponential random graph model to analyze the formation mechanism and evolution law of the social network of all 335 students in the college.(2)A method for predicting student performance based on social network is proposed.This thesis transforms the problem of student performance prediction into a problem of classifying students according to their academic performance.The obtained student social network is used as the input of the graph,and the basic characteristics of the students are selected through the random forest,and finally the student social network and the selected characteristics are input into the Graph Convolutional Network(GCN)model.The results show that after the introduction of student social network information,the accuracy of predicting student performance has been significantly improved.This thesis has made an attempt to explore the causal relationship and its application of college students' social behaviors,which can help managers better understand college students' social relationships and provide references for better education and management of students.