3Summary (1136612), страница 2
Текст из файла (страница 2)
More educated people often use theInternet for capital-enhancing activities (such as job seeking, seeking for medical or professionaladvice, and acquiring new knowledge) while lower educated people use it predominantly forentertainment [37, 38]. It was also found that students with low socioeconomic status enroll lessoften in online courses and if they are enrolled then they successfully complete it less often thantheir peers [39, 40].In most cases, the comparison is made between broad socio-demographic groups (suchcharacteristics as gender, level of education, etc. are analyzed [41]). The information about the useof the Internet is obtained from surveys. In our work, we suggest using much more detailed datanamely the information about the academic performance of students and detailed information abouttheir actual behavior on a social networking site.6Aim and objectives of the studyThe aim of the study is to develop methods for mining and analyzing VK data and to applythese methods to understand the nature of the relationship between students' behavior on the SNSand their academic achievements.
To achieve this goal, the following objectives were set:— To develop a methodology for extracting reliable data from VKontakte and combining itwith educational data; identify possible biases in the sample of VKontakte users.— To study the structure of students’ online ties and its dependence on the structure ofeducational organizations; to analyze the structure of online ties between students of differenteducational organizations and its dependence on the geographical distance between them.— To determine whether there is a relationship between students’ online ties and theiracademic performance, and to study the evolution of this relationship over time.— To identify and analyze the relationship between the online interests of students and theiracademic performance.7Methodology and design of the studyFor the purposes of the study, we developed a software program that performs queries to theVKontakte API and receives a list of all users of a certain age who indicated that they are studyingat a given educational organization.
The program matches and merges information from userprofiles with information about students from the educational organization. After the merge, thedata is anonymized and stored for later research.A direct matching by first and last name allows identifying only a small proportion of thestudents. That is why we additionally created an extensive dictionary of various forms of the samename (for example, Ivan-Vanya-Vanyusha), and also developed an algorithm for matching namesthat are written in Latin and Cyrillic.
The program searches not only among users who indicatedthat they are studying in a given educational organization but also among their friends on the socialnetwork.In the case when information about students was not available, we excluded from the samplethose users who did not have friends from their own educational organization. This approach allowsone to efficiently filter out profiles with inaccurate information.We use this method to collect four unique data sets that combine educational data with dataon the online behavior of students.— “School”, a complete sample of students from a Moscow school from the 5th to 11thgrades (N = 766)— “University”, a complete sample of students from HSE University (N = 15 757)— “City”, a complete sample of users who indicated that they are studying or havegraduated from one of Saint Petersburg’s schools (N = 1 742 392)— “Country”, a representative Russian sample of students who were born in 1995–1997 (N= 4 893)The latest sample includes participants of a longitudinal panel study “Trajectories ineducation and career” who agreed to provide their personal information for research purposes [42].To study the relationship between the structure of friendship ties and the structure ofeducational organizations, we compute modularity Q for partitions of the entire networkcorresponding to the partition of a school into grades and buildings and the partition of a universityinto years of studies, educational programs, and campuses.
To ensure that observed values of Qcannot be explained by random noise, we use a permutation test. We fixed the structure of anetwork and randomly permute the attributes of nodes (i.e. school grades, educational programs,8etc.). We then computed Qrand for such a random network. Comparison of 10 000 generated Qrandvalues with Q allows us to compute the significance level for the observed results.
To study therelationship between the structure of friendship ties with geographical distances, we analyze therelationship between the probability of the friendship tie between two schools and the distancebetween them.To investigate the detailed evolution of social ties within an educational organization, weuse information about the interaction between students instead of information about their friendshipties. We split the period of the study into 3-month intervals and then for each of the intervals webuilt a network of social ties.
We assume that the tie between two users exists if one of them givesat least one “Like” to another on VKontakte during this period. We then compute the Pearsoncorrelation coefficient between the academic performance of students and the average academicperformance of students connected to them. This approach allows us not only to determine the levelof differentiation of social ties by academic performance but also to track its evolution with time.The level of social ties differentiation on a city scale was estimated by computing acorrelation coefficient between the average USE scores of graduates of a school and the averageUSE scores of graduates from schools with which this school is connected on the SNS.
Two schoolsare considered connected on SNSs if there is at least one friendship tie between their students onsuch sites. To ensure that any observed effect is not driven by geographical locations of schools, weuse a random graph model that preserves the relationship between the probability of a friendship tiebetween schools and geographical distance between them.We analyze the interests of VKontakte users using information about their subscriptions tovarious public pages (groups) on the SNS. We compute the average GPA for subscribers todifferent pages along with their average age and proportion of females among them. To show thatthere is a correlation between interests and academic performance, we build a model that canpredict academic performance from information about subscriptions.
We identify the maincomponents containing information about students’ interests and then use them in a linearregression model to predict academic performance.9Main resultsReliability of VKontakte dataOur approach to extracting information about the online behavior of students and combiningit with educational data [43] allows us to significantly increase the total coverage of students incomparison to the direct matching of profiles (Table 1). We identify 88% of students from the“School” sample, 93% of students from the “University” sample and 82% of students from the“Country” sample.
We do not find any significant differences between identified and not identifiedstudents from the “School” sample in academic performance, age, or gender (Tables 2, 3). Similarresults are obtained for the “University” sample [44].Table 1. Percentage of students whose VK profiles were identified using the proposedmethods.Dictionary of first names with alternativeformsFriends listNoYesNo18%27%Yes57%88%Table 2. Percentages of identified VK users who did not indicate their school and/or usedalternative forms of their names, by age (grade).Grade5th6th7th8th9th10th11thPercentage of students identified85%89%88%90%88%91%85%Percentage of students who did notindicate their school64%72%69%74%70%58%72%Percentage of students who usedalternative forms of their names39%36%29%33%33%31%38%10Table 3. Groups of school students differing in the way of presenting their personal data onVK, by gender and academic performance.Girls ratioGPAIdentified on VK46%3.80Not identified on VK48%3.79Those who did not indicate their school48%3.77Those who used alternative forms of their names50%3.79For the “Country” sample, the proportion of identified users does not depend on city size(Table 4), with the only exception of rural-type settlements where the proportion of females issignificantly lower.Information about friendship ties allows to effectively filter out fake profiles.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
