Content uploaded by Anna Sigridur Islind
Author content
All content in this area was uploaded by Anna Sigridur Islind on Oct 22, 2021
Content may be subject to copyright.
Exploring study profiles of Computer Science students with Social Network
Analysis
Nidia Guadalupe L´
opez Flores
Department of Computer Science,
Reykjav´
ık University, Iceland
nidia20@ru.is
Anna Sigridur Islind
Department of Computer Science,
Reykjav´
ık University, Iceland
islind@ru.is
Mar´
ıa ´
Oskarsd´
ottir
Department of Computer Science,
Reykjav´
ık University, Iceland
mariaoskars@ru.is
Abstract
Information technology is widely adapted in all
levels of education. The extensive information resources
facilitate enhanced human capacity and the social
environment to support learning. In particular, Social
Network Analysis (SNA) has been broadly used in
teaching and learning practices. In this paper, we
perform community detection analysis to identify the
learning behavior profiles of undergraduate computer
science students in a Nordic university. The social
network was created using 273 responses to an online
survey. The students themselves provided their social
connections at the university, and node attributes were
created based on responses to questions regarding
Educational Values, Goals Orientation, Self-efficacy,
and the university teaching methods. We analyze
the biggest communities to identify the factors that
characterize the learning strategy and preferences of
undergraduate computer science students.
1. Introduction
Information technology has become an essential tool
of education. It is rich in information resources and can
extend human capacity and the social environment to
support learning. As part of its rapid growth, Social
Network Analysis (SNA) has been broadly used in
teaching and learning practices [1]. Recent literature
suggests that the potential of Learning Analytics and
Educational Data Mining offer benefits through the
use of educational data both for teachers and students
to further understand the way students approach their
learning [2]. With the increase in online learning
brought on by the COVID-19 pandemic, there is
a greater need for understanding students’ social
structures in relation to study preferences and motivation
so that universities can better accommodate the needs of
more students, especially of underrepresented students
[3].
This paper analyzes and describes the study profile
of undergraduate students of four Computer Science
related programs through Social Network Analysis and
community detection. In particular, we are interested
in knowing their learning preferences regarding group
working, physical attendance of lectures, self-efficacy
perception and goal orientation. We aim to answer
the research question: Can Community Detection
algorithms applied to the social network of students
identify undergraduate study profiles at a Computer
Science Department? To answer the question, we
analyzed the social connections of undergraduate
students and their study preferences to outline their
study profiles. The social network is built based on
an online survey created and distributed in 2019 to 717
undergraduate computer science students to investigate
their learning patterns and behaviors. The students’
social network is assortative and has a high clustering,
common features in social networks. We discover
five communities of students, where each of them is
characterized by a different study profile.
The rest of this paper is organized as follows. In
the next section, we discuss related research on learning
analytics, educational data mining, social network
analysis and study profiles. In Section 3 we present
the methodology used in this research followed by
the results in Section 4. The paper concludes with a
discussion on the implications and limitations of our
work and directions for future work.
2. Related work
2.1. Learning Analytics and Educational Data
Mining
Learning Analytics (LA) and Educational Data
Mining (EDM) have emerged as impactful research
fields that draw on educational data in the last decades
[4]. LA is defined as the “measurement, collection,
analysis and reporting of data about learners and
their contexts, for purposes of understanding and
optimizing learning and the environments in which it
occurs” [4]. The implementation of LA strategies
has been highlighted among the priorities of higher
education institutions [5]. However, EDM is defined
as “an emerging discipline, concerned with developing
methods for exploring the unique types of data that come
from educational settings, and using those methods
to better understand students, and the settings which
they learn in” [6], it employs data mining theories
and techniques to analyze educational data. Both
LA and EDM aim to improve and create methods
that enhance education at all levels. They revolve
around personalization, adaptive learning, predictive
analysis and user behavior profiling [7, 8]. Furthermore,
LA and EDM methods had been widely applied
to address a large set of concerns, e.g., predicting
students’ performance, retention analytics, intelligent
feedback provision and course recommendation [9].
Notwithstanding, limited research on learning styles
like the personalization of learning, learning style
identification and its application in teaching, learner
motivation, and student profiling has been carried out
[6].
Among the data mining techniques used in EDM,
prediction methods, like classification and regression,
and structure discovery methods, like clustering and
factor analysis, are the most commonly used. Most
recent research in EDM had been focused on the use
of two or more methods [10]. Usual clustering and
classification problems in EDM and LA can be extended
to Social Network Analysis [11, 12]. However, Social
Network Analysis has been used less frequently to
examine educational data [10].
2.2. Network Science in educational context
Among SNA applications in the educational context,
recent research has been focused on a wide range of
areas of interest. Homophily is a fundamental property
of social networks; it establishes that people with similar
properties are more likely to connect [13]. Nguyen
et al. [14] analyzed homophily regarding gender,
ethnic minority identity, family income, and academic
performance using WIFI log data. Their studies
confirmed homophily concerning demographics and
academic performance and showed that gender-based
homophily increases over time. Shirvani et al. [15]
applied SNA modeling techniques to analyze the social
dimension and learners’ roles on MOOCs discussion
forums and their changes over time. Their research
found that activity level can be predicted one week in
advance based on the course structure, forum activity
and properties of the communication network.
Community detection is one of the most significant
problems in Social Network Analysis; the analysis
of closely linked social groups is one of the
comprehensible methods of describing social structures
[11]. In higher education, community detection
algorithms have been successfully implemented
to address varied topics about learning processes.
Sturlud´
ottir et al. [16] identified fields of interest
in the courses offered in undergraduate education
programs. Xu et al. [17] analyzed discussion forum
data of MOOCs courses to gain insights on the creation
of social structures and how they change over time.
Finally, Yassine et al. [18] used community detection
algorithms to study users’ engagement patterns on
online learning networks.
2.3. Learning style, study patterns, and study
profile
Learning style theories are used in an educational
context to improve learners’ learning strengths and
instructors’ teaching abilities. EDM methods had been
used to investigate learning styles [6]. Ahmad and Tasir
[19] used log files of online learning activities to analyze
the behavior patterns of engineering students; they
concluded that the course structure, students’ previous
experience, and subject influence the thresholds defined
for learning style identification. Costaguta and Menini
[20] studied the relationship between learning style and
performance to improve group creation. More recently,
Shobbrook et al. [21] implemented elements of EDM to
validate the Fedler and Silverman’s Index of Learning
Styles (ILS) developed for engineering education. In
their research, no correlation supporting the validity of
the ILS was found, except for Lecture attendance. The
research about learning styles had been controversial
due to the limitations in measuring and determining the
learners’ learning styles individually [22].
EDM methods are also applied to investigate study
patterns in varied contexts. Shirvani et al. [23]
research presented a data-driven approach to identify
and trace study patterns in an unsupervised manner
and a hypothesis-driven approach to extract predefined
patterns from learners’ interactions. Casey and Azcona
[24] used the student activity pattern for early detection
of poor performers and to identify topics that the
students found less interesting or more difficult to
understand. Regarding using SNA to investigate study
patterns, Leeet al. [25] analyzed click-stream data using
hierarchical clustering analysis to identify behavior
patterns concerning the use of a video discussion
platform. They analyzed the transition pattern between
consecutive activities in a video discussion platform.
Considering the difficulties related to the individual
identification of learning styles, this paper applies
Social Network Analysis and community detection
methods to analyze the undergraduate students’ study
profiles. For our purposes, we define the study profiles
through a set of attributes related to educational values,
goals, self-efficacy perception, and teaching methods
preferences. Our approach relies on homophily to
explore the characterization of the study profiles by
analyzing the structure of the student community instead
of focusing on the individual characteristics of each
student.
3. Methods
3.1. Dataset
The data in this study were collected from
an online survey distributed to 717 undergraduate
students enrolled in the four computer science bachelor
programs at Reykjav´
ık University; BSc Computer
Science, BSc Computer Science research-based, BSc
Software Engineering, and BSc Discrete Mathematics
and Computer Science. The survey was created
and distributed in 2019, before the pandemic. It
included 42 questions related to Institutional Support,
Educational Values, Goals, Self-efficacy and Academic
apathy, based on a students’ readiness survey, the
Academic Readiness Questionnaire [26]. The survey
was initially distributed to study and understand
undergraduate students’ learning patterns and behaviors.
The Cronbach’s alpha coefficient of this questionnaire in
this sample is 0.70, indicating good internal consistency
reliability [27].
Among the 42 questions included in the survey, the
first two questions asked for age and gender. Questions
three to five were related to Institutional support; the
students were asked about the amount of information
they had regarding the university, their degree program
and their career possibilities. In questions six to eight,
the students’ Educational values were evaluated; to
measure the priority degree assigned to the university
studies and the grades obtained. The following ten
questions were Goal-related; in this section, the students
were asked about their drivers for goal definition,
organization, learning behavior, methodological
preferences, and long-term expectations. The following
twelve questions addressed the student’s Self-efficacy
perception; expected performance, skills and abilities,
self-motivation, confidence and capacity of adaptation
were addressed in those questions. In the next section,
Academic apathy was measured in four questions
asking for the student’s effort, work avoidance, and
scheduling level when planning their study sessions.
The following seven questions were related to the
university teaching methods, their preferences about
attendance to lectures and practical sessions, as well
as honors achievements. The last three questions were
about the students’ willingness to work in groups and
their social network size. Except for age, gender and
honors achievement, the questions’ answers were on a 5
point Likert scale. Additionally, the students were asked
to provide the list of students they most communicate
with at the university; the maximum length of the
list was 10. The survey’s response rate was 38%,
with 273 students answering it. Nonetheless, among
those responses, only 218 students provided a list of
connections at the university.
In the light of the COVID-19 pandemic and the
sudden change to Emergency Remote Teaching (ERT)
[28], and in line with recent research developed to
analyze and understand its impact on teaching and
learning processes, we decided to use the data obtained
from the survey previously implemented to analyze
the preferred study style of Computer Science students
before the pandemic, to understand the impact of ERT
in the undergraduate student community. Questions
in Table 1 were selected to perform pre-pandemic
study profile identification; these questions relate to
the self-perception of motivation, adaptation to different
teaching styles, preference to attend or not to lectures,
and their willingness to work in groups instead of
working alone. The list of students with whom the
respondent communicated at the university was used
to build a social network of students by creating a
link between the respondent and everyone that they
named. The data pre-processing was performed in R
and R-studio, while the Social Network Analysis and
community detection were performed in Python with
NetworkX.
3.2. The Girvan-Newman Algorithm
The Girvan-Newman algorithm was used to identify
the communities in the student network. This algorithm
successively removes the edges with the highest
betweenness as those edges tend to connect different
clusters [11]. Betweenness is a centrality measure
helpful to identify the most influential people in a social
network. To calculate it, the times a node (edge) is
crossed by the shortest path between any other pair
of nodes in the network are quantified. The higher
the betweenness coefficient, the more essential the
node (edge) is to connect with the rest of the network
[29]. The Girvan-Newman algorithm returns a set of
partitions where each of them represents the clusters
identified from the connected components after each
edge is removed. As there is no natural benchmark
Table 1. Questions selected to perform study profile identification
Topic Question/Statement
Individual Background Age in years
Individual Background Gender
Educational Values Getting good grades is important to me
Goals I’m a very methodical person
Self-Efficacy I can easily adapt to different styles of teaching
Self-Efficacy I can motivate myself to study when I need to
University teaching methods I like the way of teaching (the methods) used at the university
University teaching methods I do not usually attend lectures at the university
University teaching methods I watch the lectures online, on Echo360 in Canvas, rather than attend class
University teaching methods I always attend problem solving classes
Social Networks I prefer to work in groups (arranged by the teacher), rather than work on my own
Social Networks I prefer to work in groups (chosen by students), rather than work on my own
for the identified clusters, each partition returned by the
algorithm was evaluated in its modularity to select the
partition that maximizes it. The modularity coefficient
compares the edges among nodes in the same cluster and
the edges among nodes belonging to different clusters
[13].
4. Results
4.1. Network Description
Nodes in the network represent a student who
either answered the survey or was mentioned by
someone who did. Directed edges were created from
the student (source) who mentions another student
(target). Questions’ answer values were included as
node attributes in the network. The final network
displayed in Figure 1 includes 615 nodes with 806
edges. About 22% of the students who answered the
survey did not provide a list of friends at the university.
Those students, 59 in total, are included in the network
as singletons representing 9.5% of the total nodes in
the network. There are several reasons explaining the
singletons: (1) the students do not have connections in
the university, (2) the students do not feel comfortable
sharing information about their connections, or (3) as
the survey was not mandatory, the student skipped the
last part of the questionnaire.
The density and the clustering coefficient are
measures commonly used to describe the structure of
a network. The density is defined as the fraction
of connected nodes among all the possible pairs in
the network. With a maximum value of 1, the
higher the value, the more connected the network is
[13]. The average clustering coefficient measures, on
average, the extent to which the neighbors of each
node in the network link to each other [30]. Both
measures are helpful to outline characteristics of the
Figure 1. Friendship Network of undergraduate
students.
network, such as its completeness and connectedness.
The friendship network constructed has a density of
0.00213, and its average clustering coefficient is 0.1213.
Networks with low density are told to be sparse;
real-world networks are commonly characterized by
sparsity [13]. Real-world networks with comparable
densities coefficients were found in yeast protein
interactions (2,277 links and a density of 0.001)
and US air transportation data (18, 617 links and a
density of 0.004) [13]. An assortative network is
defined by [31] as networks with ”a preference for
high-degree vertices to attach to other high-degree
vertices”. The assortativity coefficient of a network
is calculated as the correlation among the degrees
of each pair of nodes in the network. Networks
with positive coefficients are known as assortative,
whereas negative values lead to disassortative networks
[13]. With an assortativity coefficient of 0.24, we can
say this network is assortative; students with many
connections tend to frequent other students with a high
number of connections. The network has 94 connected
components, the biggest with 358 nodes, while the
smaller ones are singletons.
Figure 2. In-degree distribution.
Figure 3. Out-degree distribution.
Figures 2 and 3 display the in-degree and out-degree
distribution of the nodes in the network. The in-degree
value represents the number of times the student appears
in others’ list of connections, whereas the out-degree
is the number of friends or connections declared by
each student. The in-degree distribution is right-tailed;
most students have a low in-degree (are mentioned
by fewer people) than the maximum in-degree of the
network who has an in-degree of 8. For the out-degree
distribution, most of the nodes have an out-degree of
zero. The reason for that is, among the 615 nodes
in the network: (1) only 273 answered the survey,
those nodes that were mentioned by someone but did
not answer the survey will have out-degree zero; (2)
among those who answered, 59 were singletons with
no connections. Figure 4 displays the distribution when
the zero out-degree nodes are not considered. It is
remarkable the number of nodes with a high out-degree.
Figure 4. Out-degree distribution for degrees greater
than zero.
4.2. Community Detection
The directed network was transformed into an
undirected network before applying the community
detection algorithms. With this change, the final
amount of edges decreased to 739, and the average
degree of the nodes is 2.40. Figure 5 displays the
modularity coefficient for each of the partitions returned
by the Girvan-Newman algorithm, the partition with
the highest modularity, 0.89, has 110 communities.
The number of communities is high due to the
59 singletons in the network; each singleton is
a single community. We analyze the attributes
of the five largest communities and present their
preferred study profile. The communities were named
‘Star pupils’, ‘Independent students’, ‘Team players’,
‘Female power’, and ‘Versatile students’. In those
communities, 162 students are included. Figure 6
displays them colored by the community they belong
to. ‘Star pupils’ is the only community disconnected
from the others. In the following subsections, the
study profile that characterizes the communities will be
outlined based on the distribution of the answers to the
attributes in the Academic Readiness Questionnaire [26]
in Table 1. Figures 7 and 8 display the average and
median of the responses to each question by converting
the Likert scale into: Strongly Agree=5, Agree=4,
Neutral=3, Disagree=2,and Strongly Disagree=1.
4.2.1. Community No. 1: ‘Star pupils’. The first
community identified includes 35 students. Among
them, 69% are males, 20% female and 11% unknown.
The response rate of the students in this community is
37%.The study profile is characterized by:
• Its members are mostly younger than 22 years.
Figure 5. Modularity coefficient in each step of the
Girvan-Newman algorithm.
Figure 6. The five biggest communities identified
using the Girvan-Newman algorithm.
• They declare that getting good grades is important
to them.
• They are highly methodical, and they can highly
adapt to different styles of teaching.
• They also declare being always able to motivate
themselves when needed.
• These students also say they do like the teaching
methods at the university.
• They always attend lectures, and they prefer to
attend rather than to watch recordings. They
always attend practical classes.
• These students prefer to work alone rather than
working in groups arranged by the teacher.
4.2.2. Community No. 2: ‘Independent Students’.
This community includes 32 students. The gender
distribution is quite different from the first community
analyzed; 56% males, 38% females, and 6% unknown.
In this community, the response rate was 38%. The
study profile of this community is featured by:
• Most of the students are around 25 years old, but
the ages range from 23 to 38.
• Getting good grades is important, but they are not
very methodical.
• They declare they could adapt to different
teaching styles, but they mostly like the teaching
methods at the university.
• What makes this community special is that they
do prefer to watch the lecture recordings.
• Finally, this community prefers to work in groups
chosen by themselves rather than working alone.
4.2.3. Community No. 3: ‘Team players’. This
community consists of 38 students. A 63% of them are
males, 20% females, and 8% unknown. Similar to the
previous communities, its response rate is 40%. Among
its features is possible to identify:
• The students are around 23 years old, with ages
from 20 to 26 years.
• Getting good grades is important, and they
declared themselves to some extent methodical.
• They agree they can adapt to different teaching
methods, and they mostly like those used at the
university.
• The students usually attend lectures at the
university and also problem-solving classes.
• In contrast to the other communities, these
students do prefer to work in groups chosen by
themselves rather than working alone.
4.2.4. Community No. 4: ‘Female Power’. This
community has 29 students. It is the only community
with a higher percentage of females, 55%, whereas
34% are males and 10% unknown. Additionally,
this community has the lowest response rate among
the five communities analyzed, 24%. The features
characterizing this community are:
• Their age goes from 20 up to 30 years with a
uniform distribution.
• In this community, getting good grades is
essential.
• They declare they could adapt to different
teaching methods.
• In addition, they could prefer to work alone rather
than in groups chosen by the teacher, but also
prefer to work in groups rather than alone.
Figure 7. Average responses of the communities based on the Likert scale: Strongly Agree (5), Agree (4),
Neutral (3), Disagree (2), Strongly Disagree (1).
4.2.5. Community No. 5: ‘Versatile Students’.
Twenty-eight students are allocated to this community.
It is the community with the highest percentage of
males, 82%, while 7% are females, and 11%unknown.
The response rate was 42%. This community shares
most of its features with the previous communities
presented:
• Its members are around 27 years old.
• Grades are important, and they declare to be able
to adapt to different styles of teaching.
• They declare themselves to be able to motivate
themselves when needed and mostly like the
teaching methods at the university.
• These students attend lectures almost always.
• Regarding group work, they prefer to work in
groups chosen by themselves, or alone if the
groups are chosen by the teacher.
4.2.6. Singletons. There are 59 singletons in the
network, most of them have less than 30 years. The
gender distribution is similar to the distribution of the
whole network; almost 65% are males. Among their
features; getting good grades is important to them, but
there is no evident definition of being methodical when
studying; they could adapt to different styles of teaching,
motivate themselves when needed to, and they like the
teaching methods at the university; regarding attendance
to lectures, the distribution of the answers is uniform
among the statements, but most of them agree on
prefer watching the lecture recordings instead; finally,
regarding group working, there is no clear definition
when the teacher chooses the groups, but they tend to
slightly prefer to work alone rather than groups chosen
by the students themselves.
4.3. Statistical comparison of attributes’
distribution
Differences among the distribution of the
communities’ attributes were evaluated using the
Kruskal-Wallis test. Only four attributes showed
differences in their distributions considered statistically
significant; (i) Question 4 in Table 1: Being methodical,
(ii) Question 5 in Table 1: Adaptation to teaching
styles, (iii) Question 8 in Table 1: Lecture attendance,
and (iv) Question 9 in Table 1: Watching recorded
lectures. Among those attributes, the community
‘Star pupils’ has significant differences in Being
methodical for ‘Independent students’, ‘Versatile
students’, and ‘Female power’ communities. Regarding
Lecture attendance and Watching recorded lectures, a
significant difference was found between ‘Star pupils’
and ‘Independent students’. In contrast, for Adaptation
to teaching styles, a significant distribution difference
was found between ‘Star pupils’ and ‘Female power’.
5. Discussion & Conclusion
This study presents five different study profiles
among the undergraduate students in the Computer
Science department at Reykjav´
ık University. The data
were gathered from an online survey distributed in 2019
as an initial approach to understanding undergraduate
Figure 8. Median of the communities’ responses based on the Likert scale: Strongly Agree (5), Agree (4),
Neutral (3), Disagree (2), Strongly Disagree (1).
students’ learning patterns and behaviors. Relevant
features of the student network in this university before
the pandemic are; students with many connections
tend to interact with other students who also have
many connections, but also, the network has a high
percentage of singletons, students without connections.
As the second step in this analysis, the Girvan-Newman
algorithm was used to identify the communities; the
optimal partition was selected based on its modularity.
The largest communities were analyzed to identify the
features and learning preferences that characterize the
study profile of its members. In the third section,
we identified five communities with an evident and
particular profile. (1) ‘Star pupils’ is featured by
being those who always behave as expected and have
many of the best attributes. Being the community
with the youngest members (mostly less than 22
years), first-year students are likely allocated to this
community and keeping most of the study habits they
used to have at high school. (2) With students slightly
older than the first community presented (around 23),
‘Team players’ members consider the group work
an essential factor in their learning preferences, as
long as they can choose their groups. (3) The
third community, ‘Independent students’, has students
around 25 years. This community is featured by
preferring lecture recordings instead of always being
at the university’s venues. (4) ‘Versatile Students’
community has, on average, the oldest students (around
27 years). This community share features with the
previous communities. From the first four communities,
we can infer that the year of study and maturity level
play an important role in determining the study style
profile and preferences. As the students move on in
their undergraduate studies, they become able to adapt
their learning strategy to the needs and requirements
of each course, presumably more complex in the last
terms of their studies, becoming a ‘Versatile Student’.
Last but not least, the fifth community, ‘Female Power’,
is characterized by being the only community with
more females than males. In line with the results
obtained by [14] gender-based homophily is present.
Nevertheless, besides gender, the attribute that features
this community is how important getting good grades
is. Regarding singletons, most of them prefer lecture
recording, and as it could be expected, they prefer
working alone; if the teacher or the students choose the
groups, it does not matter.
The sample of the students examined in this
paper falls under STEM, which stands for Science,
Technology, Engineering and Mathematics. Within
the literature on STEM, there has been an ongoing
discussion on the issues related to students’ sense of
belonging. On that note, an extensive body of literature
has focused on solutions targeted towards developing
an increased sense of belonging, which is thought to
lower the impact of identity-related issues on education;
one such identity-related issue can be due to skewed
distribution between genders in STEM [32]. This
paper is a contribution to that literature through the five
communities.
Among the limitations of this study, the data was
gathered with an online survey. It does not allow linking
the students with the average grade, the number of
credits earned, or the year of study. That information
could be helpful to analyze how the communities
evolve through the years, to what extent the students
interact with and provide support to peers from other
years, and how the study style selection relates to
academic performance. Other drawbacks of this data
collection method are the (i) response rate [33], affecting
the performance of the Kruskal-Wallis tests due to
small group sizes and (ii) response biases related to
social desirability and the tendency to always select
extreme ends in the Likert scale [34]. Regarding the
sample used in this analysis, the students surveyed
belong to the Computer Science department; students
in other departments may have different study styles,
so the identified communities’ presence should not be
generalized to students in other departments. This
paper focused on the biggest communities; 30% of the
students with at least one connection are part of the
communities and profiles presented. Therefore, the
other communities should be analyzed to identify if their
study profiles are similar to those identified.
The results regarding the profiles of undergraduate
students lead to relevant implications and future work.
Before the pandemic arose, only one of the five
identified study profiles preferred lecture recordings
over the university venues. The students in the
community ‘Independent students’ could have fewer
difficulties during the ERT as they had previously
interacted with the lecture recording platform. In
contrast, most students were used to attending
face-to-face lectures, and their transition to distance
learning could be more complicated. Furthermore, the
impact of the forced distance learning period could
be bigger for the youngest students, ‘Star pupils’, as
they highly prefer university venues. Another affected
community was ‘Team players’, who highly valued
interacting with their peers; during the pandemic,
their interactions were limited due to the pandemic
restrictions to contain the spread of the virus. Recent
studies on the effects of the pandemic in education
provide insights into how the new normality in Higher
Education will be [35, 36, 37, 3]. On one side, the
use of the technology and the transition to hybrid
learning are expected to be present in the new normality
of higher education [36]. In contrast, the impact of
ERT on the students’ experience with online learning
may modify their preferences about enrolling in online
courses in the future [35, 38]. The communities and
study profiles analyzed in this study correspond to the
pre-pandemic period. Consequently, once the pandemic
finishes, it is essential to study and understand the
changes in the learning profile and preferences of the
students in the communities highly affected by the
ERT. Furthermore, understanding to what extent the
pre-pandemic communities’ structure remained or not
after the pandemic will allow redesigning teaching
strategies to provide better support to the students
according to their specific profiles.
The past term, Spring 2021, ran still with restrictions
due to the COVID-19 pandemic. Most of the schools
around the world remained closed. The term Autumn
2021 will be, maybe, the first term of Education
adaptation after the pandemic. Future work of
this research relates to analyzing the undergraduate
students’ new connections, study profiles, self-efficacy
perception, and goals during the adaptation to a new
normality in higher education. In addition, other
university departments and other data sources, such as
data generated from LMSs or external tools, such as
forum activity or recorded lectures, will be included
to enrich the SNA and student communities’ profiling.
The integration of those data sources will allow linking
the results obtained in this study with similar research
performed using SNA on LMSs data, making our
conclusions more generalizable. Finally, more research
is also needed in investigating the evolution of students’
social networks through their years of study at the
university and how their modifications relate to their
study profile and performance.
References
[1] R. Huang, J. M. Spector, and J. Yang, Educational
Technology. Springer, 2019.
[2] A. Nguyen, L. Gardner, and D. Sheridan, A Design
Methodology for Learning Analytics Information
Systems: Informing Learning Analytics Development
with Learning Design. 2020.
[3] N. G. L ´
opez Flores, A. S. Islind, and M. ´
Oskarsd´
ottir,
“Effects of the covid-19 pandemic on learning and
teaching: a case study from higher education,” arXiv
preprint arXiv:2105.01432, 2021.
[4] D. Gaˇ
sevi´
c, S. Dawson, and G. Siemens, “Let’s
not forget: Learning analytics are about learning,”
TechTrends, vol. 59, no. 1, p. 64–71, 2015.
[5] M. Gaebel, T. Zhang, H. Stoeber, and A. Morrisroe,
“Digitally enhanced learning and teaching in european
higher education institutions,” tech. rep., European
University Association absl., 2021.
[6] A. Dutt, M. A. Ismail, and T. Herawan, “A systematic
review on educational data mining,” IEEE Access, vol. 5,
2017.
[7] A. Pe˜
na-Ayala, “Learning analytics: A glance of
evolution, status, and trends according to a proposed
taxonomy,” WIREs Data Mining and Knowledge
Discovery, vol. 8, May 2018.
[8] M. Brown, “Learning analytics: Moving from concept to
practice,” 2012.
[9] S. Roy and S. N. Singh, “Emerging trends in
applications of big data in educational data mining and
learning analytics,” in 2017 7th International Conference
on Cloud Computing, Data Science Engineering -
Confluence, pp. 193–198, 2017.
[10] A. Aleem and M. M. Gore, “Educational data mining
methods: A survey,” in 2020 IEEE 9th International
Conference on Communication Systems and Network
Technologies (CSNT), pp. 182–188, 2020.
[11] C. C. Aggarwal, Data Mining. Springer International
Publishing, 2015.
[12] G. Deeva, S. Willermark, A. S. Islind, and
M. Oskarsdottir, “Introduction to the minitrack on
learning analytics,” 2021.
[13] F. Menczer, S. Fortunato, and C. A. Davis, A first course
in network science. Cambridge University Press, 1 ed.,
2020.
[14] Q. Nguyen, O. Poquet, C. Brooks, and W. Li, “Exploring
homophily in demographics and academic performance
using spatial-temporal student networks,” in Proceedings
of the 13th International Conference on Educational
Data Mining, EDM 2020, Fully virtual conference, July
10-13, 2020 (A. N. Rafferty, J. Whitehill, C. Romero,
and V. Cavalli-Sforza, eds.), International Educational
Data Mining Society, 2020.
[15] M. S. Boroujeni, T. Hecking, H. U. Hoppe, and
P. Dillenbourg, “Dynamics of mooc discussion forums,”
in Proceedings of the Seventh International Learning
Analytics & Knowledge Conference, LAK ’17, (New
York, NY, USA), p. 128–137, Association for Computing
Machinery, 2017.
[16] E. G. Sturlud ´
ottir, E. Arnard´
ottir, G. Hj´
almt´
ysson, and
M. ´
Oskarsd´
ottir, “Gaining insights on student course
selection in higher education with community detection,”
2021.
[17] Y. Xu, C. Lynch, and T. Barnes, “How many friends can
you make in a week?: evolving social relationships in
moocs over time,” in EDM, 2018.
[18] S. Yassine, S. Kadry, and M.-A. Sicilia, “Application of
community detection algorithms on learning networks.
the case of khan academy repository,” Computer
Applications in Engineering Education, vol. 29, no. 2,
pp. 411–424, 2020.
[19] N. Ahmad and Z. Tasir, “Threshold value in automatic
learning style detection,” Procedia - Social and
Behavioral Sciences, vol. 97, p. 346–352, 2013.
[20] R. Costaguta and M. de los Angeles Menini, “An
assistant agent for group formation in cscl based on
student learning styles,” in Proceedings of the 7th Euro
American Conference on Telematics and Information
Systems, EATIS ’14, (New York, NY, USA), Association
for Computing Machinery, 2014.
[21] R. Shobbrook, P. Branch, and P. Ling, “Using
educational data mining to test the validity of learning
style theory,” in 2020 IEEE International Conference
on Teaching, Assessment, and Learning for Engineering
(TALE), pp. 490–496, 2020.
[22] P. A. Kirschner, “Stop propagating the learning styles
myth,” Computers & Education, vol. 106, p. 166–171,
2017.
[23] M. Shirvani Boroujeni and P. Dillenbourg, “Discovery
and temporal analysis of mooc study patterns,” Journal
of Learning Analytics, vol. 6, no. 1, p. 16–33, 2019.
[24] K. Casey and D. Azcona, “Utilizing student activity
patterns to predict performance,” International Journal
of Educational Technology in Higher Education, vol. 14,
no. 1, 2017.
[25] S. Y. Lee, H. S. Chae, and G. Natriello, “Identifying
user engagement patterns in an online video discussion
platform,” in EDM, 2018.
[26] J.-C. Lemmens et al.,Students’ readiness for university
education. PhD thesis, University of Pretoria, 2011.
[27] K. S. Taber, “The use of cronbach’s alpha when
developing and reporting research instruments in science
education,” Research in Science Education, vol. 48, Dec
2018.
[28] C. Hodges, S. Moore, B. Lockee, T. Trust, and A. Bund,
“The difference between emergency remote teaching and
online learning,” 2020.
[29] L. Igual and S. Segu´
ı, Introduction to Data Science.
Springer International Publishing, 2017.
[30] A.-L. Barab´
asi and M. P´
osfai, Network science.
Cambridge: Cambridge University Press, 2016.
[31] M. E. J. Newman, “Assortative mixing in networks,”
Physical Review Letters, vol. 89, no. 20, 2002.
[32] R. Robnett, “The role of peer support for girls and
women in stem: implications for identity and anticipated
retention,” International Journal of Gender, Science and
Technology, vol. 5, no. 3, pp. 232–253, 2013.
[33] B. Blumberg, D. R. Cooper, and P. S. Schindler, Business
research methods. McGraw-Hill Higher Education,
3 ed., 2011.
[34] Handbook of Learning Analytics. Society for Learning
Analytics Research (SoLAR), May 2017.
[35] E. M. Aucejo, J. French, M. P. Ugalde Araya, and
B. Zafar, “The impact of covid-19 on student experiences
and expectations: Evidence from a survey,” Journal of
Public Economics, vol. 191, p. 104271, 2020.
[36] A. Benito, K. Dogan Yenisey, K. Khanna, M. F. Masis,
R. M. Monge, M. A. Tugtan, L. D. Vega Araya,
and R. Vig, “Changes that should remain in higher
education post covid-19: A mixed-methods analysis of
the experiences at three universities,” Higher Learning
Research Communications, vol. 11, no. 0, 2021.
[37] S. Roy and B. Covelli, “Covid-19 induced transition
from classroom to online mid semester: Case study on
faculty and students’ preferences and opinions,” Higher
Learning Research Communications, vol. 11, no. 0,
2020.
[38] K. O’Neill, N. Lopes, J. Nesbit, S. Reinhardt, and
K. Jayasundera, “Modeling undergraduates’ selection of
course modality: A large sample, multi-discipline study,”
The Internet and Higher Education, vol. 48, p. 100776,
2021.
