Campus Behavior Information Network Based Similar Student Search And Behavior Prediction

With the release of the national standard "Smart Campus Overall Framework" in 2018,a smart campus dedicated to building an integrated campus work,study and life is gradually taking shape in many universities across the country,and the educational concept "From Course to Life" has been widely accepted.The traditional education mode based on pre-made teaching plans can no longer satisfy the current personalized cultivation needs of innovative talents.With the rapid development of new generation of information technology,in recent years,researchers have explored novel ways to improve the quality of talent cultivation by utilizing the student data,such as applying big data analysis to discover subtle but meaningful information as the guidance for better student management.Lifestyle is a comprehensive manifestation of students’ psychological status,financial status and hobbies,as well as has an important impact on students’ personal development and academic performance.Searching for students with similar lifestyles and predicting student behavior are very important for caring about students’ mental health,financial status,and academic performance.However,most of the existing studies are not appropriate for campus behavior data and also have not been solved well to some problems.In the similar student search,they cannot explain the similarities between students and fail to dynamically integrate more data sources.In the task of student behavior prediction,they cannot obtain the temporal and periodic features of the data.The lifestyles and behavior patterns of students in the campus environment are recorded in the campus behavior data,which can be represented by the network(or graph)structure.The concept of campus behavior information network is used to model campus behavior data.The campus behavior information network is not a definite structure,because different research tasks need to reflect the features of the data from different perspectives.By using campus behavior data,we have the following contributions on the work of similar student search and student behavior prediction:(1)Analyzing and summarizing the dense,multi-source,dynamic and periodic features of campus behavior data.As for the problem of similar student search,the structure of multi-source campus behavior information network is designed,and the problem definition of similar student search is proposed.In response to the needs for student behavior prediction problem,the dynamic campus behavior information network based on hypergraph is designed and define the student behavior prediction task.(2)Aiming at the similar student search problem in the multi-source campus behavior information network,the SCALE(similar campus lifestyle student miner)algorithm is designed and implemented.A constrained meta-path similarity calculation method PathSimC is proposed to adapt to the dense features of campus behavior data.We support the expansion of data sources by sequentially constructing student similarity subnetwork and student similarity networks to express similarities under multiple data sources and semantics.The algorithm finally obtains the representations of students through network representation learning for similar student search task.At the same time,each part of the SCALE algorithm is decoupled and a parallelization strategy is designed to improve efficiency.Finally,the real-world datasets collected in the campus environment are used to verify the effectiveness and execution efficiency of the SCALE.(3)Facing the student behavior prediction problem in the dynamic campus behavior information network,the SCAN(student campus behavior prediction)algorithm is designed and implemented.SCAN utilizes the GRUS module to obtain the temporal features in the dynamic campus behavior information network.For extracting the periodic feature in the campus behavior data,SCAN innovatively employs the GRUP module to update the representation of the hidden layers and nodes at periodic intervals and design the optimization goals applicable to the SCAN.Meanwhile,a variety of feasible framework design strategies are designed and discussed,and the rationality of the SCAN algorithm framework is analyzed.Finally,the real campus datasets are used to verify the effectiveness of the SCAN and the rationality of the framework design.