| Big data analysis tracks many social processes,especially in educational scenarios that receive widespread attention from the general public.A thorough and in-depth understanding of the complex phenomena that emerge in educational settings contributes to providing data support for educational practices and offering scientific references for education administrators in implementing management decisions.It holds significant academic and practical value.Investigating the interaction between individuals and the external environment during the educational process is not only a core issue in the pedagogy itself but also has attracted widespread attention across various disciplines including computer science,complexity science,network science,psychology,sociology,and economics.Traditional researches in pedagogy faced limitations due to constraints in data sources and analytical tools.In terms of data,there were issues of small scale,limited timeliness,and difficulty in eliminating social desirability biases.In terms of methodology,traditional statistical analysis methods struggled to accurately capture the complexity that emerged from the interaction between individuals and the external environment,and they also lacked predictive capabilities for future changes.With the high-speed advancement of information technology and the advent of the big data era,unprecedented opportunities have emerged for studying the interaction between individuals and complex external environments.The diversity of data acquisition methods and the richness of data types have significantly enhanced data accessibility.The accumulation of large-scale,high-quality data has facilitated methodological transformations and tool innovations in analyzing the interactions between individuals and the external environment.Under new opportunities,as the studies of leveraging empirical methods to understand education gradually accumulates,an interdisciplinary research branch known as computational pedagogy has emerged,focusing on the utilize of large-scale actual data and quantitative methods to analyze various complex phenomena and issues in computational settings.This dissertation focuses on the crucial role of analyzing students’ behavioral data,exploring the methods and applications of computational pedagogy in higher education settings through large-scale empirical data analysis at three levels in the order of scope from local to global,saying individual behaviors,interactions,and collective behaviors.The contents and main innovation points of this dissertation are summarized as follows.At the level of individual behavior analysis,based on unobtrusive behavioral data,this dissertation examines and quantifies the regularity of student behaviors and the predictive power in relation to academic performance.Firstly,from the unobtrusive on-campus daily behaviors generated by approximately 28 million smart campus card digital records of nearly twenty thousand undergraduates,two types of high-level behavioral characters were extracted based on the time binning method: one type represents the regularity of daily life behavior,referred to as the orderliness measure; the other type represents the level of study effort,referred to as the diligence measure.Then,an entropy-based metric,namely,actual entropy,was applied to characterize and distinguish the degree of orderliness behaviors among different individuals.Next,analyzing the correlation between the orderliness measure,diligence measure,and academic performance separately,it was found that the orderliness measure has its independent effects on academic performance.In particular,orderliness is for the first time,to our knowledge,proposed as an important behavioral character that is not directly related to study behavior but significantly and positively correlated with academic performance.Finally,leveraging the Rank Net model and the two high-level behavioral characters,academic performance prediction was conducted.It was found that introducing the orderliness measure in the prediction model improved the accuracy of the predictions.At the level of interaction analysis,based on the null model method in the field of complex systems,this dissertation explores the mutual influence of students’ academic performance,revealing the extent of peer effects and its affecting factors.Firstly,based on the data characteristics of random assignment of over 5,000 undergraduates in Chinesestyle campus dormitories,the dissertation investigated peer effects within dorm rooms.Then,the academic performance was divided into different tiers.It was found that the tier combination of students in the dorm room had a significant difference between the actual probability and the theoretical probability.Additionally,there was a phenomenon of convergence in roommates’ academic performance.Next,a measure named assimilation was proposed to characterize the degree of similarity in roommates’ academic performance.By combining the assimilation measure and null model techniques,the existence of peer effects in dorm rooms was demonstrated and the strength was quantified.To our knowledge,null model method was for the first time presented as the measure to quantify the peer effects in Chinese-style on-campus dorm rooms.It was manifested as a significantly higher level of similarity in academic performance among roommates compared to random chance,in which the difference reaches 10.7%.Furthermore,by leveraging regression models,an analysis was conducted to identify factors influencing academic performance in dorm rooms.It was found that students’ academic performance was influenced by higher-order factors such as roommates’ average performance,roommates’ heterogeneity,and ordinal rank in dorm rooms.Finally,null model techniques and falsification tests were employed together to demonstrate the significance of these factors in influencing academic performance.At the level of collective behavior analysis,based on the mechanism that individuals’ preferences are embedded in their social behaviors,this dissertation established student social networks from complex and collective temporal-spatial interaction data.In addition,this dissertation investigated the structural characteristics of student social networks and the association with students’ mental health status.Firstly,based on students’ longitudinal check-in time-spatial data at on-campus dining establishments over 21 weeks,this dissertation utilized a spatial-temporal inference method to construct offline social networks,namely,student co-occurrence networks.Then,the topology changes of the student cooccurrence networks were analyzed,revealing the association with students’ interpersonal interactions at different time periods.Finally,by accessing the relationship between the node centrality in the co-occurrence networks and the flourishing trait within the scope of positive mental well-being,it was discovered that there was a significant positive correlation between node centrality and the improvement of the flourishing trait within the student co-occurrence networks.In addition,the correlation exhbited an overall increasing trend.Computational pedagogy is a vibrant interdisciplinary research branch that faces both opportunities and challenges in the dimension of behavioral data analysis.In order of effect scope from local to global,this dissertation utilizes three large-scale empirical datasets to demonstrate the application of students’ behavioral data analysis at three levels: individual behaviors,interactions,and collective behaviors.In terms of methodology,the research paradigm that combines large-scale unobtrusive behavioral data with interdisciplinary analysis tools will inevitably become the mainstream in the field of behavioral data analysis. |
PDF Full Text Request