The Study Of The Elimination Of The Homophily Effect In Students’ Social Network Mining

The ever-developing information technology facilitates communication and cooperation between people.For college students,good social relations not only create a good learning and living atmosphere,reduce their mental stress,but also improve their competitiveness in society.For educators,they can have a clearer grasp of students by comprehending the social situation of students.They can then improve their management methods based on such knowledge.Therefore,mining social relations of college students is of great significance to the development of college students’physical and mental health and college student’s management.However,how to extract the social relationship of college students is still one of the problems in the current information education research.Fortunately,educational big data techniques have made it possible to extract hidden college students’ social relationship.The campus card records the daily transaction behavior data of students on campus,which implies the social behavior and social relationships of students.At present,researchers have studied the problem of mining college students’social networks from the data recorded by campus cards.Some mining algorithms are proposed based on the association between behavioral "co-occurrence" and social relationships.However,most of the research ignores the influence of homophily(major homophily,grade homophily,etc.)on mining college students’ social network,which makes cross-group social relationships difficult to be explored.Aiming at these problems,this thesis proposes a social relationship mining method based on the sliding time-window method and a hierarchical co-occurrence model.We also verify the effectiveness of the model and analyze some interesting properties of the social networks and nodes in the network.The main contributions and innovations of this thesis are as follows:(1)Using the sliding time-window method,the co-occurrence is extracted from the transaction record and the number of co-occurrences is calculated.This solves the problem of erroneous omission of co-occurrence in the fixed time slice method.(2)A hierarchical co-occurrence model based on association analysis is proposed to eliminate the impact of homophily on social relationship mining.This model uses the association analysis and "quasi-association analysis" to mine social relationships of students within the same group(with homophily)or not within,thus eliminating the influence of homophily.(3)An adaptive method for determining the thresholds of the social relationship of each student is proposed.This method tackles the difficulty in setting the thresholds and establishes a theoretical foundation.(4)By comparing the social networks of mining,the validity of the model is verified.Results show that the inter-group social relationship mining ability is significantly improved.We also analyze the node and network characteristics of the extracted social networks and obtain some social behavior pattern of college students.