Content uploaded by Vitomir Kovanovic
Author content
All content in this area was uploaded by Vitomir Kovanovic on Sep 26, 2016
Content may be subject to copyright.
Content uploaded by Srecko Joksimovic
Author content
All content in this area was uploaded by Srecko Joksimovic on Mar 18, 2015
Content may be subject to copyright.
What do cMOOC participants talk about in Social Media?
A Topic Analysis of Discourse in a cMOOC
Srećko Joksimović
School of Interactive Arts and
Technology
Simon Fraser University
Burnaby, Canada
sjoksimo@sfu.ca
Amal Zouaq
Department of Mathematics and
Computer Science
Royal Military College of Canada
Kingston, Canada
amal.zouaq@rmc.ca
Vitomir Kovanović
School of Informatics
The University of Edinburgh
Edinburgh, UK
v.kovanovic@ed.ac.uk
Dragan Gašević
Schools of Education and Informatics
The University of Edinburgh
Edinburgh, UK
dgasevic@acm.org
Jelena Jovanović
Department of Software Engineering
Faculty of Organizational Sciences
Belgrade, Serbia
jeljov@fon.rs
Marek Hatala
School of Interactive Arts and
Technology
Simon Fraser University
Burnaby, Canada
mhatala@sfu.ca
ABSTRACT
Creating meaning from a wide variety of available information
and being able to choose what to learn are highly relevant skills
for learning in a connectivist setting. In this work, various
approaches have been utilized to gain insights into learning
processes occurring within a network of learners and understand
the factors that shape learners’ interests and the topics to which
learners devote a significant attention. This study combines
different methods to develop a scalable analytic approach for a
comprehensive analysis of learners’ discourse in a connectivist
massive open online course (cMOOC). By linking techniques for
semantic annotation and graph analysis with a qualitative analysis
of learner-generated discourse, we examined how social media
platforms (blogs, Twitter, and Facebook) and course
recommendations influence content creation and topics discussed
within a cMOOC. Our findings indicate that learners tend to focus
on several prominent topics that emerge very quickly in the
course. They maintain that focus, with some exceptions,
throughout the course, regardless of readings suggested by the
instructor. Moreover, the topics discussed across different social
media differ, which can likely be attributed to the affordances of
different media. Finally, our results indicate a relatively low level
of cohesion in the topics discussed which might be an indicator of
a diversity of the conceptual coverage discussed by the course
participants.
Categories and Subject Descriptors
J.1 [Administrative Data Processing] Education; K.3.1 [Computer
Uses in Education] Distance learning
General Terms
Human Factors, Algorithms
Keywords
Connectivism, Content analysis, SNA, cMOOC
1. INTRODUCTION
The initial development of Massive Open Online Courses
(MOOCs) dates back to 2005, and coincides with the ideas of
connectivism and networked learning [1]. While the first publicly
available MOOC was the Connectivism and Connective
Knowledge (CCK08) course in 2008, it was in 2011 when
MOOCs started gaining significant attention [2]. Although
MOOCs very quickly became an important component of the
adult online education, there is presently an extensive debate
about their role in higher education [3, 4]. The main concerns are
related to the effective scaling-up of traditional courses and the
ability of MOOCs and their underlying pedagogy to meet the
needs of higher education [3].
Within the last several years, two prominent types of MOOCs
evolved. The more centralized type of MOOCs – xMOOCs – are
focused on content delivery to large audiences, where the learning
process is teacher-centered, i.e., based on transferring knowledge
from instructors to learners [5]. xMOOCs are usually delivered
using a single platform (learning management system), where
learners receive knowledge (most commonly in a video format),
and further apply that knowledge in projects defined by the
teacher [5]. On the other side of the spectrum, more distributed
MOOCs emerged (cMOOCs). In cMOOCs, teachers’ role is
primarily focused on the early instructional design and
facilitation. cMOOCs do not rely on any centralized platform but
rather use various social media for sharing information and
resources among learners. The main goal of learning in cMOOCs
is knowledge building through connection and collaboration with
peers [6]. Learners are co-creators of the content and there is no
formal evaluation of the learning achievements.
The most commonly indicated issues and challenges related to
MOOCs are low course completion rates, high degree of learner
attrition, and the lack of a theoretical framework that would allow
for better understanding of learning processes in networked
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for
components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to
post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from Permissions@acm.org.
LAK '15, March 16 - 20, 2015, Poughkeepsie, NY, USA
Copyright 2015 ACM 978-1-4503-3417-4/15/03…$15.00
http://dx.doi.org/10.1145/2723576.2723609
156
learning [7]. In their analysis of the research proposals submitted
to the MOOC Research Initiative1 (MRI), [7] showed a promising
upturn in addressing a wide variety of the challenges recognized
to date. Majority of submissions proposed well-established
frameworks in educational research and social sciences as a
foundation for examining and understanding learner motivation,
metacognitive skills, and other factors that shape learning and
teaching in MOOCs.
However, our literature review indicates that most of the current
studies on cMOOCs are based on quantitative methods and rather
simple metrics (e.g., the frequency of facilitators’ and learners’
postings) [8, 9]. Without the capacity to explain practice and
complexity of networked learning, existing approaches and
research models do not allow for understanding of learning at
scale [10]. To contribute to the current research practices in this
area, our study proposes a combined use of automated content
analysis and social network analysis (SNA) in order to provide a
more effective approach to MOOC research. More precisely, the
study reported in this paper suggests an analytic method that
integrates quantitative (automated content analysis and SNA) and
qualitative analysis of posts created within different social media
platforms used in a cMOOC. Relying on tools for automated
concepts extraction, as well as SNA tools and techniques, we were
able to identify main groups of concepts emerging from learners’
posts and to analyze how they evolve throughout the course.
Further qualitative analysis enabled a more in-depth interpretation
of our findings.
Having that cMOOCs often incorporate various technologies into
the learning process, our first objective was to examine how
different social media influence the discourse of course
participants. The second objective was related to the role of
course facilitators in a cMOOC. More precisely, our objective was
to analyze how course readings, suggested by course facilitators,
frame the topics being discussed among learners. Finally, we were
interested in analyzing learners’ discourse through a temporal
dimension, that is, how topics discussed by students changed over
time, when certain topics emerged and whether we can identify
topics that sustained throughout the course.
2. THEORETICAL BACKGROUND AND
RESEARCH QUESTIONS
2.1 Connectivism and cMOOCs
The theoretical foundation behind cMOOCs is connectivism [1,
11] and its principles of autonomy, diversity, openness and
interactivity [12]. Connectivism is proposed as a novel theory of
learning for “the digital age” [13]. It assumes abundance of
information and digital networks, and views learning as the
development and maintenance of networks of information,
resources and contacts [14]. Primary activities in connectivist
learning are [12]: i) aggregation, ii) remixing, iii) repurposing, and
iv) forwarding of resources and knowledge.
Teaching in connectivist setting differs from common practices in
distance and online education. In particular, teaching is focused
on instructional design and learner facilitation, while the course
content is created by course participants (i.e., learners and
facilitators) [5, 6]. Kop et al. [15] therefore argue that the key to
cMOOC success is a combination of teaching and social presence
that enables an effective facilitation of learners’ self-regulation of
learning, which in turn leads learners to the accomplishment of
worthwhile, personalized and authentic learning outcomes.
Instead of being a distant “rock star” academic of xMOOCs [16]
[p. 58], a teacher in cMOOC is expected to be a role model [14],
1 http://www.moocresearch.com
and a discussion moderator rather than a tutor [12]. According to
Kop et al. [15], instructors are “aggregating, curating, amplifying,
modeling, and persistently being present in coaching or
mentoring. The facilitator also needs to be dynamic and change
throughout the course“[p. 89]. For this delegation of content
creation from the instructor to the network, Yaeger et al. [9]
emphasize the need for a strong core of active participants that
would provide the critical mass of activity.
A typical design of a cMOOC assumes collaboration between
course participants using various social media (e.g., blogs,
Twitter, Facebook, Google+, RSS feeds and mailing lists) [17].
The use of particular tools and their affordances can directly
influence and support the community formation [18], which is
essential for learning within cMOOC environments. Twitter
hashtags are probably the best example of technological
affordances that can affect community formation [19]. However,
the abundance and diversity of technology in cMOOCs is also a
challenge [20]–[22], and a source of potential disconnect between
the sub-communities in the course [14]. For example, a study by
Mackness et al. [21] found that variations in the level of expertise
and use of different platforms lead to the development of sub-
communities which reduced possibilities for autonomy, openness
and diversity. While cMOOC literature acknowledges the
importance of technology for shaping learning experience, the
effects of particular technologies are rarely discussed [3].
The cMOOC literature so far has mainly focused on descriptive
methods for research and analysis of learning in a networked
environment. Perhaps, the most comprehensive approach was
applied in the study of Fournier et al. [23], who relied on counts
of contributions/posts (e.g., Moodle discussion blogs, Twitter),
survey, virtual ethnography, discourse analysis and educational
data mining, in order to describe learning processes in the PLENK
cMOOC. However, their discourse analysis relied on manual
coding of messages, a highly time consuming process, while the
quantitative methods applied (i.e., clustering and correlational
analysis) did not provide a more detailed insight into the
underlying learning processes. Although studies by Kop [9], and
Yeager et al. [20] adopted social network analysis, the application
was limited to the illustration of interactions within the course
discussions. Finally, Wen et al.’s [24] study on discourse centric
learning analyzed the association between learners’ discourse and
attrition in a MOOC, using the Latent Dirichlet allocation
approach. However, they did not consider the principles of
connectivism, nor did they consider different social media
platforms.
2.2 Research questions
While the number of studies about MOOCs is growing [25], there
have been very few studies that looked into the effects of
particular choices of technology on shaping learning in cMOOCs.
The exceptions are studies by Fini [17] and Mak et al. [26].
However, they primarily focused on quantitative analysis of
interactions, media affordances and learning approaches, which
did not provide insights into the content of learners’ discussions.
In our study, we wanted to examine learners’ discourse in
different social media that are typically used in cMOOCs – i.e.,
Facebook, Blogs and Twitter. The main objective was to obtain an
insight into the topics that learners mentioned in their posts, and
how these topics differ across different media. Accordingly we
defined our first research question as follows:
RQ1: Do topics discussed by learners differ across social media
used in a cMOOC?
In such a dynamic environment, where learners are encouraged to
choose what they want to learn and make sense of the high
volume of available information through sustained collaboration
157
with other learners in a network, we were interested in examining
the role of facilitators in shaping the discussions in the course.
While the study by Skrypnyk et al. [27] identified the key role of a
small number of active facilitators and technological affordances
in shaping the information flow and formation of interest-based
communities, it is still an open question how much these
communities remain within the original course curriculum
suggested by the instructors. Given that cMOOCs are typically
organized as a series of online events led by respected facilitators
in a particular domain [15], it seems reasonable to analyze how
much influence those facilitators have on shaping the overall
discussion between learners. This is likely related to the level of
autonomy of learners, their self-regulation of learning, and their
particular learning goals. Therefore, we defined our second
research question:
RQ2: To what extent do the readings suggested by the course
facilitators shape the topics discussed by learners in social media
in a cMOOC?
We were also interested in examining whether the discussed
topics stabilize over time or perhaps change in accordance with
the changes in the course’s weekly topics. This led us to our third
research question:
RQ3: How do topics discussed by learners change over time in a
cMOOC across different social media?
Finally, we aimed at providing a scalable approach for a
comprehensive analysis of learners’ discourse in cMOOCs. The
study by Skrypnyk et al. [27] examined the use of particular
Twitter hashtags over time and thus, to some extent examined the
content of learner messages and their evolution over time. Still,
our study provides a more comprehensive coverage of learners’
generated discourse by investigating blog posts, Twitter messages
and Facebook discussion messages.
3. METHODOLOGY
3.1 Study context
To get a better insight into the emerging topics in a cMOOC and
answer our research questions (RQ1-3), we analyzed the content
created and exchanged through social media in the scope of the
2011 installment of the Connectivism and Connective Knowledge
(CCK11) cMOOC (http://cck11.mooc.ca/). The CCK11 course
was facilitated through 12 weeks (January 17th – April 11th 2011),
with the aim of exploring the ideas of connectivism and
connective knowledge, and examining the applicability of
connectivism in theories of teaching and learning. The topics
covered throughout the course included: i) What is
Connectivism?, ii) Patterns of Connectivity, iii) Connective
Knowledge, iv) What Makes Connectivism Unique? v) Groups,
Networks and Collectives, vi) Personal Learning Environments
and Networks, vii) Complex Adaptive Systems, viii) Power and
Authority, ix) Openness and Transparency, x) Net Pedagogy: The
Role of the Educator, xi) Research and Analytics, and xii)
Changing Views, Changing Systems. The course participants
were provided with readings recommended by the course
facilitators for each theme covered by the course (one theme per
week). The facilitators encouraged learners to “remix” and share
their new knowledge through various means including blogs,
Twitter and Facebook2. The participants were also provided with
daily newsletters that aggregated the content they created and
exchanged through these blogs, tweets and Facebook posts.
Content aggregation was done using gRSShoper. Finally, the
2 A complete list of the instructions provided to CCK11 participants is available at
http://cck11.mooc.ca/how.htm
course included weekly live sessions that were carried out using
Elluminate.
3.2 Data Collection and Analysis
The overall process of data collection and analysis was done in
several steps that are outlined below.
Collection of learners’ posts and recommended readings. We
relied on gRSShopper to automatically collect blog posts and
tweets, while Facebook posts were obtained using the official
Facebook API3. All posts were stored in a JSON format for
further processing. Table 1 provides descriptive statistics of the
posts collected. Besides posts, we also collected readings
recommended by the course facilitators for each theme covered by
the course. The recommended readings appeared in the course
outline4 for each week of the course.
Semantic annotation of learners’ posts and recommended
readings. Having collected learners’ posts and recommended
readings, the next step was to semantically annotate them, i.e., to
associate their content with concepts that reflect the semantics of
those posts and readings. To this end, we examined and tested
several state-of-the-art semantic annotation tools, including
TagMe5, WikipediaMiner6, Alchemy API7, and TextRazor8.
Based on the analysis of the annotations produced by the
examined tools on a sample of the collected posts, and also based
on the previous examinations of these tools reported in the
literature (e.g., [28-30]), we made the following decision: short
posts (tweets and Facebook messages) were annotated using
TagMe, while Alchemy API was used for the annotation of longer
posts (i.e., blog posts) and recommended readings. Both tools
annotate content with Wikipedia concepts which made all the
annotations consistent (i.e., based on the same concept scheme).
Since today’s annotators mostly operate on English texts, we
made use of a freely available language translation tool (Microsoft
Translation API9) to translate non English posts (5% of our
dataset) to English. Even though the resulting translations were
not ideal, in most cases, we noticed that they preserved the gist of
the original content.
Having inspected the annotations of posts and readings, we
identified certain invalid concepts originating from the
imperfection of today’s semantic annotators. To reduce a potential
negative impact on further analysis, we manually removed all
concepts that were obviously erroneous (e.g., concept ‘cable
television’ was identified as a disambiguation of the term
‘networks’, or ‘environmentalism’ was associated with ‘[learning]
Table 1. Descriptive statistics of the collected data: number of
active learners, post counts (total, average, SD), and word
count for each media analyzed
Media Active
participants
Post
count
Average post
count (SD)
Word
count
Blog 193 1473 3.13 (4.80) 428626
Facebook 78 1755 5.03 (5.23) 67883
Twitter 835 2483 1.80 (3.85) 43180
Total 997 5711 - 539689
3 https://developers.facebook.com
4 http://cck11.mooc.ca/outline.htm
5 http://tagme.di.unipi.it/
6 http://wikipedia-miner.cms.waikato.ac.nz/
7 http://www.alchemyapi.com/products/alchemylanguage/concept-tagging/
8 http://www.textrazor. com/
9 http://msdn.microsoft.com/en-us/library/dd576287.aspx
158
environments’), as well as concepts that could not be considered
valid in the context of our analysis (e.g., Lady Gaga’s songs).
Once we created a list of erroneous concepts, the removal was
done automatically – before including a concept, we would ensure
that the concept is not specified within the list.
Creation of concept co-occurrence graphs. The extracted
concepts served as an input for the creation of undirected
weighted graphs for each week of the course and each media
analyzed (36 graphs in total). Aiming to identify the most
important concepts and their connections, we created graphs
based on the co-occurrence of concepts within a single post. For
example, if concepts C1 and C2 appeared within the same post,
the two concepts were included in a graph as nodes and the edge
C1-C2 was created. Each edge was assigned a weight representing
the frequency of co-occurrence of the two concepts.
Clustering of concepts into topics (concept clusters). To further
analyze relationships between concepts in the constructed graphs,
and extract clusters of concepts, we applied a modularity
algorithm for community detection [31]. The initial analysis
revealed a rather high number of clusters (over 50 on average, in
case of Twitter graphs), with very few large groups and a
significant number of small clusters (individual concepts or pairs
of concepts). Therefore, we decided to extract the largest
connected component in each graph, and use these components
for cluster detection [36–38]. The size of the largest connected
components used in the study varied from 88% to the size of the
total graph in case of blogs, from 78% to 94% in case of
Facebook, and from 52% to 86% of the total graph size in case of
graphs extracted from Twitter.
In order to better understand emerging topics (i.e., clusters of
concepts), we performed an in-depth qualitative analysis. We
initially examined concepts within each cluster, aiming to reveal
potential patterns that would provide description for the cluster
analyzed. In cases where such a pattern could not be revealed, we
focused on the content of the messages that these concepts were
extracted from, to provide a better context for our interpretation.
Computation of graph metrics. The constructed graphs were
analyzed using graph metrics that are commonly used for analysis
of collocation networks [35]:
Graph density – the ratio of existing edges to the total
number of possible edges,
Weighted cluster density – for each of the clusters we first
calculated its graph density, and then calculated weighted
average cluster density, where weights are cluster sizes.
Radius – the minimum eccentricity among all nodes,
Diameter – the maximum distance between two nodes,
Network centrality measures, namely weighted degree (the
count of edges a node has in a network, pondered by the
weight of each edge) and betweenness centrality (the
indicator of node’s centrality in a graph).
The first three metrics were used to measure the level of
coupling/spread of concepts (i.e., coherence) discussed in the
analyzed posts, whereas the centrality measures served to measure
the importance of individual concepts. Specifically, higher degree
centrality should indicate concepts that are associated with many
other concepts, while higher betweenness centrality could be seen
as an indicator of concepts that could potentially “bridge” two or
more topics [36]. Moreover, the selection of these metrics was
motivated by the findings of contemporary research on automated
assessment of learner generated content and information
extraction. For example, Whitelock et al. [33] used keyword-
based graphs for automated essay assessment and automated
feedback provision. Their study showed that highly connected and
dense graphs indicate better structured essays [37]. Building
further on the research in computational linguistics, we expected
that graphs with higher density would imply a more cohesive and
coherent text [38]. Using the measure of degree, density, radius,
and diameter, we aimed at examining whether and how the use of
different media influences the “structure and cohesiveness” of the
content being generated.
Computing similarity of posts as well as posts and recommended
readings. To answer our research questions, we also needed to
examine if there were topics of pertaining interest/relevance to
learners, so that they kept discussing them even after the course
progressed to other topics. To this end, for each social media
analyzed, we computed the cosine similarity [39] between
concepts discussed in each pair of consecutive weeks (i.e.,
concepts extracted from posts in the corresponding two weeks). In
particular, we relied on a vector representation of the concepts
discussed each week, and used the cosine similarity metric to
compute similarity between concepts in two consecutive weeks.
In a similar manner, we computed similarity between concepts
discussed in posts and those discussed in recommended readings.
In this case, the readings recommended for week k, k=1..11 were
compared to posts in each succeeding week (k+1, k+2,…). The
idea was to identify learners’ interest in the course themes, based
on the assumption that learners would discuss more topics that
they find interesting/relevant.
4. RESULTS
In order to gain an initial insight into the topics discussed in each
media channel, in Figure 1 we report the number of identified
topics (i.e., concept clusters) identified and the most dominant
topics for each media and each course week (Table 2, expressed
as the percentage of the graph size, e.g., T1(45%)). We also
examined the strength of relationships between concepts within
the identified clusters (Figures 2 and 3); how concepts from
different media relate to one another (Figure 4); the dynamics of
concepts over the length of the course – whether and to what
extent they changed from week to week (Figure 5 and Table 2),
and how they relate to the recommended readings (Figure 6).
Figure 1. Topic (i.e., cluster of concepts) count per week per
media
Figure 1 shows the number of detected topics (i.e., concept
clusters) per week, for each media analyzed. Within the first half
of the course, the highest number of topics was extracted from
Facebook posts (except for week 1), while the messages
exchanged on Twitter showed the lowest number of topics
throughout the course.
Density of concept clusters for all analyzed social media follows
quite a similar pattern throughout the course (Figure 2). Aiming to
better understand the emerging concept clusters (i.e., topics), we
calculated graph density for each individual concept cluster, per
media and per week. It is interesting to note that the highest
159
density among the media was observed in the first week of the
course, for the concept clusters emerging from tweets. There are
also two peeks where density increased notably; for blogs within
the week 8, as well as by the end of the course in case of
Facebook. These phenomena are analyzed in more details in the
Discussion section.
Figure 2. Average density of concept clusters per week and
per media
Figure 3 further shows how concepts within topics (i.e., concept
clusters) were coupled in terms of graph radius and diameter. The
results show that concepts extracted from Facebook and blogs
posts were more tightly coupled than those extracted from Twitter
posts, which seems to indicate more homogeneous and related
discussions overall on these two media. As the course progressed,
concepts from tweets became more tightly coupled, while for
Facebook and blog posts, the coupling of concepts remained
approximately at the same level.
Figure 3 Radius (dotted lines) and diameter (solid line) of
concept clusters measured per week and per media.
Figure 4 describes similarities between concepts discussed in each
media. Comparison of concepts extracted from blogs and
Facebook posts yielded the highest similarity over the 12 weeks of
the course. On the other hand, concepts extracted from Twitter
and blog posts showed the highest discrepancy throughout the
course. It is also interesting to note the decline in similarity within
the week 11, for each pair of media compared.
In order to further examine the dynamics of concepts being
discussed, we calculated the similarity between concepts extracted
from posts in each pair of consecutive weeks (e.g., for week 4, we
calculated the semantic similarity of concepts from weeks 4 and
3). As a measure of semantic similarity, we calculated the cosine
similarity between vectors of concepts for each pair of
consecutive weeks. Figure 5 shows that in all media channels, the
concepts discussed by learners remained rather similar from week
to week. In case of Twitter posts, similarity between two
consecutive weeks tends to increase over time (except for weeks 8
to 10), while in case of blogs and Facebook, we were able to
observe a decrease over time.
Figure 4. Similarity of concepts discussed in different media
We also analyzed semantic similarity between concepts extracted
from posts exchanged on each media and recommended readings
for i) the same week, and ii) all the previous weeks. For example,
for week 7, we calculated similarity between concepts extracted
from blogs, Facebook and Twitter in week 7, and concepts
extracted from readings recommended in weeks 1 to 7. This
analysis revealed a quite consistent pattern over the three media.
Figure 6 shows that concepts extracted for each week, within all
three media, were the most similar to the readings assigned for
weeks 1-3, and 9. On the other hand, based on the extracted
concepts, readings assigned for weeks 4 to 8 had the lowest
similarity with posts from any of the course weeks. Moreover,
among the three media analyzed, results show that Twitter posts
(i.e., concepts extracted from Twitter posts) differed the most
from the content presented in the readings for each week of the
course, while blogs seemed to be the most similar to the readings.
Figure 5. Similarity of concepts discussed in two consecutive
weeks (per media)
Table 2 shows the top three topics (i.e., concept clusters) for each
media and each week. Topics are ranked based on the number of
concepts they consist of. For each topic, the table shows the top
three concepts ranked based on their betweenness and degree
centrality. Among those highly ranked concepts connectivism,
learning, e-learning, education, social media, and knowledge,
were most commonly represented within one of the three topics
for most of the weeks, within each media analyzed.
160
Figure 6. Similarity between weekly readings and posts from each week
An in-depth qualitative analysis of these results allowed us to
provide a more detailed interpretation of the topics covered within
each week, for each of the three media.
By analyzing topics identified in Twitter messages, we were able
to identify the following five groups of topics:
Within the first group of topics we recognized posts that are
related to sharing information regarding the course,
relevant publications, and other resources. These topics were
indicative of weeks 1 to 3, as well as of weeks 7 and 11.
The second group was based on topics related to
connectivism as a learning theory. It is interesting to note
that these topics were more frequent during the first four
weeks of the course. Topics in this category included
discussions on learning in networks (week 1); connectivism
and its influence on instructional design (week 2);
connectivism as one of the emerging learning theories (week
3); and unique characteristics of connectivism (week 4).
Later in the course, topics such as connectivism as a learning
pedagogy (week 8) received significant attention, as well as
the potential influence of a connectivist approach to learning
on changes in the role of instructional designers (week 9).
The third group of topics was related to the application of
connectivism in practice. The most notable points discussed
included teaching foreign languages in connectivist settings
and desirable competencies for teaching online (week 4);
necessary skills for learning in networked learning
environments (week 5); and the role of learners in
connectivism and the importance of learning analytics (week
6). The topics belonging to this group received significant
attention later in the course with the introduction of the
concept “sharing for learning” in connectivism and available
technologies for collaboration within a connectivist course
(week 9). Finally, within the week 12 the role of
connectivism in theory-informed research was also
addressed.
Within the fourth group of topics, networked learning and
establishing communities in networked learning
environments gained significant attention. Here, the course
participants were interested in topics such as taking control
of learning (weeks 2 and 3); networks and communities
emerging from MOOCs (week 3); collaboration within
networked learning environments (weeks 8 and 10); and
design and delivery of social networked learning (week 12).
The final and the largest set of topics was primarily focused on
educational technology and its application in various settings.
The most indicative topics of this group are personal learning
environments (weeks 5 and 6); social media in education (week
5); teaching with ICT and tools available (weeks 6 and 12); tools
for learning and complex adaptive systems (week 7); integration
of technological affordances into traditional classroom settings
(week 8); challenges and best practices of educating teachers to
use available technological affordances (week 9); and mobile
(week 10) and blended learning (week 11).
Our analysis of topics detected in blog posts revealed topic groups
similar to those observed in tweets, though with some observable
differences:
The first group of topics, similar to the one detected in
Twitter messages, was about sharing course resources:
information about the course and the readings (week 1), and
the concept map of connectivism (week 11).
The second group identified topics related to MOOCs in
general: the concept of MOOC, previous MOOCs (e.g.,
PLENK, CCK08) (week 1), and how MOOCs affect learning
in classroom settings (week 8). Although the topics from this
group appeared throughout other weeks of the course, these
topics were mostly discussed at the beginning of the course.
The third group of topics received significant attention within
the first five weeks of the course. This group was related to
connectivism as a learning theory, and how connectivism
relates to other learning theories. Course participants
discussed the main characteristics of connectivism (weeks 1,
4, and 12) and relationships to other learning theories (week
5); validity of connectivism as a learning theory (week 2);
teachers’ role in connectivism (weeks 3 and 8); aspects of
teaching English as a foreign language in connectivist
settings (week 5); and about collective intelligence,
constructivism, subjectivism and importance of interpretation
(weeks 5 and 10).
161
Table 2. The number of exchanged posts and three most dominant topics (with the size as a percentage of all the clusters) for each
week and each media; for each topic, the three most central concepts (sorted by betweenness and degree centrality) are given
Twitter Blogs Facebook
Week
1
Total Topics: 3 Total Posts:30
T1 (45%): concept, substantial form, social
T2 (27%): knowledge, open source, e-learning
T3 (27%): connectivism, video, constructivism
(learning theory)
Total Topics: 7 Total Posts:200
T1 (67%): learning, education, knowledge
T2 (19%): twitter, concept, teacher
T3 ( 6%): tag, critical thinking, website
Total Topics: 5 Total Posts:84
T1 (36%): connectivism, idea, learning
T2 (25%): facebook, open source, uploading
and downloading
T3 (18%): information, paradigm, twitter
Week
2
Total Topics: 7 Total Posts:270
T1 (33%): connectivism, education, e-learning
T2 (22%): employment, social network, thought
T3 (22%): learning, concept map, instructional
design
Total Topics: 7 Total Posts:159
T1 (35%): learning, knowledge, thought
T2 (18%): argument, research, computer
network
T3 (18%): motivation, facebook, MOOC
Total Topics: 11 Total Posts:260
T1 (17%): twitter, facebook, quora
T2 (17%): learning, tradition, employment
T3 (15%): education, connectivism, knowledge
Week
3
Total Topics: 6 Total Posts:256
T1 (30%): connectivism, wikipedia, conversation
T2 (26%): learning, knowledge, computer network
T3 (15%): education, e-learning, stephen downes
Total Topics: 8 Total Posts:145
T1 (19%): thought, knowledge, social network
T2 (17%): teacher, connectivism, information
T3 (17%): mind, writing, metaphor
Total Topics: 11 Total Posts:189
T1 (21%): learning, thought, connectivism
T2 (16%): linkedin, facebook, social network
T3 (11%): knowledge, idea, object (philosophy)
Week
4
Total Topics: 7 Total Posts:236
T1 (23%): connectivism, education, constructivism
(learning theory)
T2 (20%): e-learning, social network,
actor?network theory
T3 (17%): learning, information age, theory
Total Topics: 9 Total Posts:160
T1 (25%): connectivism, knowledge, social
network
T2 (24%): theory, technology, time
T3 (22%): thought, learning, education
Total Topics: 9 Total Posts:210
T1 (18%): knowledge, connectivism, social
change
T2 (18%): thought, e-learning, student
T3 (16%): learning, education, skill
Week
5
Total Topics: 6 Total Posts:271
T1 (36%): e-learning, connectivism, bonk (video
game series)
T2 (24%): edtech, internet, english as a foreign or
second language
T3 (17%): education, educational entertainment,
teacher
Total Topics: 8 Total Posts:182
T1 (27%): thought, theory, truth
T2 (20%): sound, youtube, human
T3 (18%): education, learning, connectivism
Total Topics: 8 Total Posts:269
T1 (24%): thought, knowledge, understanding
T2 (23%): learning, education, student
T3 (22%): connectivism, wiki, facebook
Week
6
Total Topics: 4 Total Posts:217
T1 (37%): connectivism, english as a foreign or
second language, behaviorism
T2 (32%): education, edtech, e-learning
T3 (21%): collaboration, knowledge, thought
Total Topics: 9 Total Posts:109
T1 (18%): learning, education, psychology
T2 (17%): feedback, connectivism, cognition
T3 (15%): theory, book, internet
Total Topics: 8 Total Posts:144
T1 (20%): learning, thought, history of personal
learning environments
T2 (18%): knowledge, information, brain
T3 (17%): diigo, blogger (service), tool
Week
7
Total Topics: 6 Total Posts:270
T1 (42%): connectivism, twitter, knowledge
T2 (24%): edtech, e-learning, mind map
T3 (14%): technology, complex adaptive system,
department of education and communities
Total Topics: 8 Total Posts:122
T1 (22%): learning, education, knowledge
T2 (17%): sense, idea, intention
T3 (14%): complexity, understanding, human
Total Topics: 6 Total Posts:73
T1 (23%): education, knowledge, culture
T2 (20%): twitter, united kingdom, facebook
T3 (18%): information, employment, history of
personal learning environments
Week
8
Total Topics: 4 Total Posts:207
T1 (37%): connectivism, writing, book
T2 (30%): education, e-learning, edtech
T3 (17%): social network, learning, power
(philosophy)
Total Topics: 4 Total Posts:71
T1 (69%): learning, social network, psychology
T2 (27%): research, neoplatonism, people
T3 ( 3%): massive open online course, internet
forum, beauty
Total Topics: 7 Total Posts:94
T1 (20%): knowledge, intelligence, information
technology
T2 (17%): education, rss, plug-in (computing)
T3 (17%): research, social media, new media
Week
9
Total Topics: 5 Total Posts:156
T1 (42%): edtech, e-learning, web 2.0
T2 (33%): internet, connectivism, file sharing
T3 (11%): learning, school, control theory
Total Topics: 9 Total Posts:87
T1 (26%): learning, education, hypothesis
T2 (22%): thought, social group, happiness
T3 (13%): skill, knowledge, literacy
Total Topics: 5 Total Posts:132
T1 (26%): education, student, technology
T2 (22%): connectivism, knowledge,
connectionism
T3 (21%): learning, thought, object
(philosophy)
Week
10
Total Topics: 5 Total Posts:160
T1 (38%): connectivism, computer network,
pedagogy
T2 (21%): e-learning, education, teacher
T3 (19%): learning, MOOC, google apps
Total Topics: 9 Total Posts:111
T1 (27%): learning, education, educational
psychology
T2 (13%): facebook, google, twitter
T3 (12%): truth, metaphor, behaviorism
Total Topics: 9 Total Posts:113
T1 (28%): learning, thought, connectivism
T2 (22%): employment, student, collaboration
T3 (19%): book, writing, child
Week
11
Total Topics: 6 Total Posts:228
T1 (36%): connectivism, social media, emergence
T2 (25%): e-learning, edtech, education
T3 (14%): learning, theory, information age
Total Topics: 7 Total Posts:76
T1 (22%): education, teacher, pedagogy
T2 (21%): learning, psychology, science
T3 (20%): thought, skill, concept map
Total Topics: 5 Total Posts:50
T1 (32%): knowledge, learning, quality
(philosophy)
T2 (21%): connectivism, thought, behaviorism
T3 (18%): value (personal and cultural),
wisdom, truth
Week
12
Total Topics: 6 Total Posts:182
T1 (31%): connectivism, web 2.0, networked
learning
T2 (28%): e-learning, education, edtech
T3 (17%): learning, english as a foreign or second
language, information age
Total Topics: 6 Total Posts:51
T1 (26%): thought, pedagogy, connectivism
T2 (24%): learning, observation, education
T3 (18%): writing, memory, attention
Total Topics: 7 Total Posts:137
T1 (22%): learning, research, connectivism
T2 (20%): google, writing, English language
T3 (18%): person, applied science, education
162
Networked learning and learning in connectivist settings
received the highest attention among the course participants
who were using blogs as a communication medium. The
main topics covered included complexity of learning in
networks, professional learning and importance of
motivation for learning in networked environments (weeks 2,
4, 7 and 12); tools for learning in networks and gathering
information (week 2); groups versus networks in connectivist
settings (week 3); importance of interactions, internal and
external feedback for learning in networks (weeks 6, 7, and
10); the source of knowledge/intelligence in networks (week
8); the role of technology in mediating teachers’ role in
networked learning (week 11), and learning affordances in
networked learning environments (week 9); and digital
literacy (week 9) and conceptual models for learning in
networks (week 12);
Discussions about online and distance education represent
the fifth group of topics. The most commonly discussed
topics included e-learning in classroom settings (week 3);
social media services and social media platforms in online
and distance education (weeks 5, 7, 8, and 10); social
networks, social groups, and emerging social communities in
distance education (weeks 6 and 9); instructional design for
alternative education (weeks 9, 10, and 12), and metrics for
measuring learners’ success in online and distance education
(week 10).
The final group of topics was concerned with educational
technology and use of ICT in education. Virtual learning
environments and their use in higher education (weeks 6 and
7), ICT for teaching foreign language (week 7), personal
learning environments (week 8) and learning management
systems in education (weeks 11 and 12), were most
commonly discussed in blog posts.
According to our analysis, learners’ messages exchanged on
Facebook remained within similar general topics:
Available resources and information about the course
content were common topics within weeks 1, 2, and 12.
Within the connectivism as a learning theory topic group,
the course participants were discussing the idea of
connectivism and its position in education (weeks 1 and 2);
how connectivism was different from the paradigm “wisdom
of crowds”, collective and connective wisdom (weeks 3 and
11); the main challenges of new learning theories (week 7);
origins of connectivism (e.g., connectivism as a connectionist
approach to learning) (week 9), and how connectivism
empowers learners to take responsibility for their learning
(week 11).
Similar to blogs, networked learning and learning in
connectivist settings received the most significant attention.
These topics were evenly distributed throughout the course,
and included networked learning and affordances that foster
learning and help development of digital literacies (weeks 1
and 2); nature of teaching and learning in connectivism
(weeks 4 and 8); social networking groups and sharing
information within networks (weeks 3, 5, and 10);
assessment in the connectivist framework (weeks 10 and 11);
and collaboration and cooperation in networks (week 11).
As with other media analyzed, educational technology was
quite significant topic starting from the week four of the
course. Institutions of higher education and their view of the
role of ICT in education (week 4); social media platforms
and connectivism (week 5); personal learning environments
and differences/similarities with learning management
systems (weeks 6 and 7); tools for collecting, sharing and
tagging resources (week 6); role of educational technology in
teaching foreign languages (weeks 9 and 10); and ICT and
intellectual ethics (week8), were the most prominent.
Opposite to blogs where topics about online and distance
education were quite prominent, within the Facebook
communication channel, topics on education in general
received more attention. Course participants were interested
in advantages and disadvantages of formal and institutional
learning (weeks 4 and 7); the role of scholars in digital
environments (week 2); how we learn and where we are
learning from (week 3); important characteristics and skills
of learners that drive learning in general, and in connectivist
settings (week 5), how to create knowledge from information
(week 6).
5. DISCUSSION
5.1 Interpretation of results with respect to
the research questions
Considering the subject of the course, it is not surprising that the
most common topics covered within each media are related to
connectivism as a learning theory, networked learning, education
(in general, and online and distance education in particular), skills
for teaching/learning in networks, and educational technology.
However, concepts discussed within each topic differ to a certain
extent. For example, among topics related to educational
technology that were discussed in blog and Facebook posts, there
was a topic covering the issues of teaching and learning with ICT.
While the course participants, who discussed this topic through
blog posts, were mostly focused on technological affordances in
teaching foreign language, posts exchanged on Facebook
discussed the same topic from the learners’ perspective.
Regarding our first research question (RQ1), we found that except
for the first week of the course and concepts extracted from
Twitter, the topics learners discussed in their posts in all three
media analyzed tended to follow a similar pattern. In particular,
posts tended to cover a wide set of concepts that quite differed
from one post to another (Figure 2). However, our findings also
indicate that concepts extracted from Twitter posts less frequently
co-occurred and were less tightly coupled within a topic than in
case of blog and Facebook posts (Figure 2 and 3). It could be
deduced that blog and Facebook allowed for writing more
coherent posts. This confirms previous findings that social media
vary in their affordances [40], in terms that certain social
platforms allow for more elaborate writing on topics of interest.
On the other hand, less coherent discourse might be an indicator
of difficulties to form a learning community. Without a clear set
of shared interest, it is unlikely that a community would emerge.
Observing though the perspective of the three media analyzed, it
seems that blogs and Facebook offer better opportunities for the
community development.
As for our second research question (RQ2), we found that posts
throughout the 12 weeks of the course mostly covered topics from
recommended readings for the first three weeks. Within those
three weeks of the course, readings included topics such as
connectivism as a learning theory, learning in networks, as well
as learning in networks and connective knowledge, which we
identified as the most common topics in the analyzed posts.
Moreover, Figure 5 shows that topics discussed within two
consecutive weeks did not differ significantly, indicating that
course participants tended to continue conversation on the topic of
interest, rather than follow new themes introduced within the
course. This suggests that those dominant themes are determined
by groups of learners who engage collaboratively, rather than by
the instructor. Therefore, we might conclude that our results
support the main theoretical assumptions of connectivism [1] and
are in line with the previous studies [8, 27]. More precisely, the
163
learning process is not focused on transferring knowledge from
the instructor to course participants, but rather on the connections
and collaboration between learners [6], while learners also
participate in content creation. Moreover Kop, et al. [15] and
Skrypnyk et al. [27] confirmed that the information flow and
knowledge building process also depend on those network-
directed learners who are willing to engage into interaction with
their peers and share knowledge among the network of learners.
Therefore, it seems reasonable to conclude that learners engage
into discussions with peers who share similar interests, thus
framing the topics discussed within each media.
Finally, regarding our third research question (RQ3), our findings
show that even though the count of topics identified within each
week changed over time and differed among the media analyzed
(Figure 1), the most dominant and high-level groups of topics
(e.g., educational technology, networked learning) quickly
emerged, and sustained throughout the course. More specialized
concepts did change in each group of topic, since learners showed
interests in various aspects of those topics (e.g., social network
analysis, personal learning environments). However, overall they
remained focused on the general groups of topics.
5.2 Limitations of this study
In order to address issues of internal and external validity of our
findings, certain limitations need to be acknowledged. The main
issues regarding internal validity originate in the process of data
collection and concept extraction. In our study, we relied on
gRSShopper for the automated collection of learners’ blog posts,
and copies of tweets. This source was used as by the time we
collected data for the study (April-August 2014), several blogs
were not available any longer. Likewise, due to the limitations
introduced by the Twitter API, we were not able to obtain original
tweets. Therefore, we turned to the posts available within the
CCK11 newsletter. Second, we relied on Alchemy API and
TagMe for the extraction of concepts from learners’ posts and
recommended readings. However, as stated in the Methodology
section, these tools produced some erroneous concepts that we
manually removed. This suggests that the extracted concepts
might not fully and correctly represent the themes discussed in
posts and readings. Finally, we relied on Microsoft Translate API
in order to translate non-English posts (5% of all the collected
posts), therefore the resulting translations depend on the quality of
the API used.
Addressing issues of external validity is important from the
perspective of generalizing our findings. Therefore, it is important
to conduct a similar analysis within a different educational
domain or course.
6. CONCLUSIONS
The reported study proposed a novel analytic approach that
integrates tools and techniques for automated content analysis and
SNA with qualitative content analysis. This approach was used for
the exploration of topics emerging from the learners’ discourse in
cMOOCs, and offered an in-depth insight into the topics being
discussed among course participants. Moreover, the proposed
analytic method also allowed for validation of certain ideas of
connectivism – e.g., learners were primarily focused on the course
topics they were interested in, regardless of the topics suggested
by the course facilitators, while the technology had a significant
impact on how learners discussed certain topics [6]. Further, our
approach might be suitable for analysis of different media used in
cMOOCs, as one of the critical features. For such multi-media
studies, it is essential to proceed to the analysis of actual content
and discourse rather than just counts of the use (e.g., page hits)
[41, 42]. This is necessary as different media have different
affordances that can affect how processes of knowledge creation
unfold in cMOOCs [18, 26].
Building a trustworthy community in diverse and large networks,
as those emerging from cMOOCs, is recognized as one of the
important challenges [26]. Being able to reveal topics discussed in
different media and among emerging social groups might help
learners to “bridge the social gap” and more easily reach groups
with similar interests. On the other hand, our study also shows an
overall low density of the analyzed concept graphs. This might be
an indicator of low cohesion among the concepts used by learners
[38], and low-to-moderate mutual understanding and consensus
built within the entire network [37]. It seems that, at the network
level, course participants could not find shared concepts of
interests within those broader topics being discussed. In addition,
our findings might indicate a lack of shared vocabulary or
conceptual models, considering that people originated from
different backgrounds and different cultures. However, a broad
consensus of the entire network – per medium – might not be
possible given the size and diversity in interests, background, and
goals of the course participants. Perhaps, a better unit of analysis
could be communities. For example, further research should
create similar graphs for specific communities – e.g., such as
those that emerged in the study reported in [27] – and analyze
their cohesion, rather than the cohesion of the entire network. We
would expect to reveal higher graph density, and more connected
graphs, as indicators of higher level of shared understanding.
Our findings also indicate that several topics gained significant
attention, while other course topics were not commonly discussed
among learners. Therefore, the question is how facilitators and/or
learners should proceed with regard to those less “interesting”
topics? Given that learners choose what to learn in cMOOCs,
should facilitators provide a better connection with those topics
that were “more popular”, or introduce “less popular” topics in
different ways, or perhaps such findings could inform the course
design, pointing out to the most important topics for the course
participants?
Further research is also needed to examine how different social
groups shape discussions and whether we can identify certain
patterns in learners’ approaches to course-related discussions,
over various social media. For example, it would be interesting to
analyze how social groups formed around certain topics evolve
over time; are there groups that use various media to collaborate
with their peers on a certain topic; and how much attention
receive topics initiated by course facilitators, compared to topics
proposed by learners.
7. REFERENCES
[1] G. Siemens, “Connectivism: A learning theory for the
digital age,” Int. J. Instr. Technol. Distance Learn., vol. 2, no. 1,
pp. 3–10, 2005.
[2] L. Pappano, “The Year of the MOOC,” The New York
Times, 02-Nov-2012.
[3] J. Daniel, “Foreword to the Special Section on Massive
Open Online Courses MOOCs–Evolution or Revolution?,”
MERLOT J. Online Learn. Teach., vol. 10, no. 1, p. 4, Mar. 2014.
[4] V. Kovanović, S. Joksimović, D. Gašević, G. Siemens,
and M. Hatala, “What public media reveals about MOOCs?,” Br.
J. Educ. Technol., vol. submitted, 2014.
[5] S. Bayne and J. Ross, “The pedagogy of the Massive
Open Online Course (MOOC): the UK view,” Curriculum
(including learning, teaching and assessment) > Pedagogic
approaches HEER000427, Mar. 2014.
[6] G. Siemens, “Teaching in Social and Technological
Networks,” Connectivism, 16-Feb-2010. .
[7] D. Gašević, V. Kovanović, S. Joksimović, and G.
164
Siemens, “Where is Research on Massive Open Online Courses
Headed? A data analysis of the MOOC Research Initiative,” Int.
Rev. Res. Open Distance Learn. IRRODL, vol. 15, no. 5, p.
accepted for publication, 2014.
[8] R. Kop, “The challenges to connectivist learning on
open online networks: Learning experiences during a massive
open online course,” Int. Rev. Res. Open Distance Learn., vol. 12,
no. 3, pp. 19–38, 2011.
[9] C. Yeager, B. Hurley-Dasgupta, and C. A. Bliss,
“cMOOCs and global learning: An authentic alternative,” J.
Asynchronous Learn. Netw., vol. 17, no. 2, pp. 133–147, 2013.
[10] M. De Laat, “Networked learning,” Police Acad. Neth.
Apeldoorn, 2006.
[11] G. Siemens, Connectivism: Learning as Network-
Creation. 2005.
[12] C. O. Rodriguez, “MOOCs and the AI-Stanford like
Courses: Two Successful and Distinct Course Formats for
Massive Open Online Courses,” Eur. J. Open Distance E-Learn.,
vol. 2012, no. II, pp. 1–13, 2012.
[13] G. Siemens, “Connectivism: A learning theory for the
digital age,” in International Journal of Instructional Technology
and Distance Learning, 2005.
[14] T. Anderson and J. Dron, “Three generations of distance
education pedagogy,” Int. Rev. Res. Open Distance Learn., vol.
12, no. 3, pp. 80–97, 2010.
[15] R. Kop, H. Fournier, and J. S. F. Mak, “A pedagogy of
abundance or a pedagogy to support human beings? Participant
support on massive open online courses,” Int. Rev. Res. Open
Distance Learn., vol. 12, no. 7, pp. 74–93, 2011.
[16] J. Ross, C. Sinclair, J. Knox, S. Bayne, and H. Macleod,
“Teacher Experiences and Academic Identity: The Missing
Components of MOOC Pedagogy,” J. Online Learn. Teach., vol.
10, no. 1, pp. 57–69, 2014.
[17] A. Fini, “The Technological Dimension of a Massive
Open Online Course: The Case of the CCK08 Course Tools,” Int.
Rev. Res. Open Distance Learn., vol. 10, no. 5, 2009.
[18] A. Gruzd, B. Wellman, and Y. Takhteyev, “Imagining
Twitter as an Imagined Community,” Am. Behav. Sci., vol. 55, no.
10, pp. 1294–1318, 2011.
[19] L. Yang, T. Sun, M. Zhang, and Q. Mei, “We know
what @you #tag: does the dual role affect hashtag adoption?,” in
Proceedings of the 21st international conference on World Wide
Web, New York, NY, USA, 2012, pp. 261–270.
[20] R. Kop, “The challenges to connectivist learning on
open online networks: Learning experiences during a massive
open online course,” Int. Rev. Res. Open Distance Learn., vol. 12,
no. 3, pp. 19–38, 2011.
[21] J. Mackness, S. Mak, and R. Williams, “The ideals and
reality of participating in a MOOC,” in Proceedings of the 7th
International Conference on Networked Learning 2010, 2010, pp.
266–275.
[22] M. Saadatmand and K. Kumpulainen, “Participants’
Perceptions of Learning and Networking in Connectivist
MOOCs,” J. Online Learn. Teach., vol. 10, no. 1, pp. 16–30,
2014.
[23] H. Fournier, R. Kop, and G. Durand, “Challenges to
Research in MOOCs,” J. Online Learn. Teach., vol. 10, no. 1, pp.
1–15, 2014.
[24] M. Wen, D. Yang, and C. P. Rose, “Linguistic
Reflections of Student Engagement in Massive Open Online
Courses,” in Eighth International AAAI Conference on Weblogs
and Social Media, 2014.
[25] T. R. Liyanagunawardena, A. A. Adams, and S. A.
Williams, “MOOCs: A systematic study of the published
literature 2008-2012,” Int. Rev. Res. Open Distance Learn., vol.
14, no. 3, pp. 202–227, 2013.
[26] S. F. J. Mak, R. Williams, and J. Mackness, “Blogs and
Forums as Communication and Learning Tools in a MOOC,”
2010, pp. 275–284.
[27] O. Skrypnyk, S. Joksimović, V. Kovanović, D. Gašević,
and S. Dawson, “Roles of course facilitators, learners, and
technology in the flow of information of a cMOOC,” Br. J. Educ.
Technol. Spec. Issue MOOCs, p. submmited for review, 2014.
[28] M. Cornolti, P. Ferragina, and M. Ciaramita, “A
Framework for Benchmarking Entity-annotation Systems,” in
Proceedings of the 22Nd International Conference on World Wide
Web, Republic and Canton of Geneva, Switzerland, 2013, pp.
249–260.
[29] L. Jean-Louis, A. Zouaq, M. Gagnon, and F. Ensan,
“An Assessment of Online Semantic Annotators for the Keyword
Extraction Task,” in Proceedings of the 13th Pacific Rim
International Conference on Artificial Intelligence, Gold Coast,
Australia, 2014.
[30] J. Jovanovic, E. Bagheri, J. Cuzzola, D. Gasevic, Z.
Jeremic, and R. Bashash, “Automated Semantic Annotation of
Textual Content,” IEEE IT Prof., vol. (to appear), 2014.
[31] M. E. Newman, “Modularity and community structure
in networks,” Proc. Natl. Acad. Sci., vol. 103, no. 23, pp. 8577–
8582, 2006.
[32] V. Batagelj and A. Mrvar, Pajek—analysis and
visualization of large networks. Springer, 2004.
[33] A. Mislove, M. Marcon, K. P. Gummadi, P. Druschel,
and B. Bhattacharjee, “Measurement and Analysis of Online
Social Networks,” in Proceedings of the 7th ACM SIGCOMM
Conference on Internet Measurement, New York, NY, USA,
2007, pp. 29–42.
[34] W. Lee, J. J. Lee, and J. Kim, “Social Network
Community Detection Using Strongly Connected Components,”
in Trends and Applications in Knowledge Discovery and Data
Mining, Springer, 2014, pp. 596–604.
[35] F. Boudin, “A comparison of centrality measures for
graph-based keyphrase extraction,” in International Joint
Conference on Natural Language Processing (IJCNLP), 2013, pp.
834–838.
[36] N. Schluter, “Centrality Measures for Non-Contextual
Graph-Based Unsupervised Single Document Keyword
Extraction,” in Proceedings of TALN 2014 (Volume 2: Short
Papers), Marseille, France, 2014, pp. 455–460.
[37] D. Whitelock, D. Field, S. Pulman, J. Richardson, and
N. van Labeke, “Designing and testing visual representations of
draft essays for higher education students,” in Discourse-Centric
Learning Analytics (DCLA) Workshop, LAK, 2014, vol. 14.
[38] A. C. Graesser, D. S. McNamara, and J. M. Kulikowich,
“Coh-Metrix providing multilevel analyses of text
characteristics,” Educ. Res., vol. 40, no. 5, pp. 223–234, 2011.
[39] A. Huang, “Similarity measures for text document
clustering,” in Proceedings of the sixth new zealand computer
science research student conference (NZCSRSC2008),
Christchurch, New Zealand, 2008, pp. 49–56.
[40] R. F. Schmid, R. M. Bernard, E. Borokhovski, R. M.
Tamim, P. C. Abrami, M. A. Surkes, C. A. Wade, and J. Woods,
“The effects of technology use in postsecondary education: A
meta-analysis of classroom applications,” Comput. Educ., vol. 72,
pp. 271–291, 2014.
[41] H. Fournier, R. Kop, and H. Sitlia, “The Value of
Learning Analytics to Networked Learning on a Personal
Learning Environment,” in Proceedings of the 1st International
Conference on Learning Analytics and Knowledge, New York,
NY, USA, 2011, pp. 104–109.
[42] P. Goodyear, Advances in Research on Networked
Learning (Computer-Supported Collaborative Learning, V. 4).
Norwell, MA, USA: Kluwer Academic Publishers, 2004.
165