Socio-Temporal Dynamics in Peer Interaction Events

Abstract

Asynchronous online discussions are broadly used to support peer interaction in online and hybrid courses. In this paper, we argue that the analysis of online peer interactions would benefit from the focus on relational events that are temporal and occur due to a range of factors. To demonstrate the possibility, we applied Relational Event Modeling (REM) to a dataset from online discussions in seven online classes. Informed by a conceptual model of social interaction in online discussions, this modeling included (a) a learner attribute capturing aspects of temporal participation, (b) social dynamics factors such as preferential attachment and reciprocity, and (c) turn- by-turn sequential patterns. Results showed that learner activity and familiarity from recent interactions affected their propensity to form ties. Turn-by-turn sequential patterns, that capture individual posting in bursts, explain how two-star network patterns form. Since two-star network patterns could further facilitate small group formation in the network, we expected the models to also capture communication in triads (i.e. triadic closure). Yet, models, devoid of the content of exchanges, did not capture the social dynamics well, and failed to predict patterns for communication across triads. By bringing in discourse features, future work can investigate the role of knowledge building behaviours in triadic closure of digital networks. This study contributes fresh insights into social interaction in online discussions, calls for attention to micro-level temporal patterns, and motivates future work to scaffold learner participation in similar contexts.

Full text

Socio-Temporal Dynamics in Peer Interaction Events
Bodong Chen
College of Education and Human Development
University of Minnesota–Twin Cities
Minneapolis, USA
chenbd@umn.edu
Oleksandra Poquet
Centre for Change and Complexity in Learning (C3L)
University of South Australia
Adelaide, SA, Australia
sasha.poquet@unisa.edu.au
ABSTRACT
Asynchronous online discussions are broadly used to support peer
interaction in online and hybrid courses. In this paper, we argue
that the analysis of online peer interactions would benet from the
focus on relational events that are temporal and occur due to a range
of factors. To demonstrate the possibility, we applied Relational
Event Modeling (REM) to a dataset from online discussions in seven
online classes. Informed by a conceptual model of social interaction
in online discussions, this modeling included (a) a learner attribute
capturing aspects of temporal participation, (b) social dynamics
factors such as preferential attachment and reciprocity, and (c) turn-
by-turn sequential patterns. Results showed that learner activity
and familiarity from recent interactions aected their propensity to
form ties. Turn-by-turn sequential patterns, that capture individual
posting in bursts, explain how two-star network patterns form.
Since two-star network patterns could further facilitate small group
formation in the network, we expected the models to also capture
communication in triads (i.e. triadic closure). Yet, models, devoid of
the content of exchanges, did not capture the social dynamics well,
and failed to predict patterns for communication across triads. By
bringing in discourse features, future work can investigate the role
of knowledge building behaviours in triadic closure of digital net-
works. This study contributes fresh insights into social interaction
in online discussions, calls for attention to micro-level temporal
patterns, and motivates future work to scaold learner participation
in similar contexts.
CCS CONCEPTS
•Human-centered computing →Social network analysis
;
Web-based interaction
;
•Information systems →
Social net-
working sites;Clustering.
KEYWORDS
Relational event modelling, temporality, digital peer networks
ACM Reference Format:
Bodong Chen and Oleksandra Poquet. 2020. Socio-Temporal Dynamics in
Peer Interaction Events. In Proceedings of the 10th International Conference
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 prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than the
author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specic permission
and/or a fee. Request permissions from permissions@acm.org.
LAK ’20, March 23–27, 2020, Frankfurt, Germany
©2020 Copyright held by the owner/author(s). Publication rights licensed to ACM.
ACM ISBN 978-1-4503-7712-6/20/03. . . $15.00
https://doi.org/10.1145/3375462.3375535
on Learning Analytics and Knowledge (LAK ’20), March 23–27, 2020, Frank-
furt, Germany. ACM, New York, NY, USA, 6 pages. https://doi.org/10.1145/
3375462.3375535
1 INTRODUCTION
Online learning frequently appears in the strategic plan of educa-
tional institutions. In higher education, an increasing number of
universities and colleges are putting their degrees online, while
hybrid courses have become the “new normal”. In K-12, the applica-
tion of online learning is also growing, not only to facilitate student
learning but also to enable richer teacher professional development
opportunities.
In online and hybrid learning experiences, asynchronous online
discussion is broadly used to support learners’ social participa-
tion and peer interaction. Online learning environments are often
equipped with asynchronous discussion forums. Instructional de-
signs also tend to incorporate online discussions to facilitate various
learning activities, such as reection on readings, question and an-
swers, and peer dialogues [
8
]. To support productive discussion
experiences, pedagogical strategies are proposed such as incorpo-
rating questioning strategies [
27
], engaging students in role-taking
[
28
], and inviting learners to take responsibility in identifying key
or promising ideas in discussion threads [
5
,
10
]. Technological in-
novations are also proposed to promote active participation and
deep engagement in online discussions. For instance, new graphi-
cal interfaces are invented to better support purposeful attention
to existing posts in online discussions[
17
]; social network sites,
such as Facebook and Edmodo, are adopted in both formal and
informal learning settings to foster peer connectedness and collabo-
ration [
24
,
29
]. Empirical studies in various contexts have reported
a positive association between discussion participation and learn-
ing performance [
11
,
12
,
31
], showing the promise of sustaining
eective discussions for learning.
In spite of prior work in these areas, the fundamental social
mechanisms that drive peer interaction in asynchronous online
discussions remain under-examined. Extant work rarely addresses
peer interaction in online discussions as a complex phenomenon
shaped by a range of factors related to learner attributes, social
dynamics, pedagogical designs, and technological constraints. This
study attempts to bridge the knowledge gap by tackling the fol-
lowing overarching question: What are the mechanisms of social
interaction in asynchronous online discussions? To this end, we
applied relational event modeling (REM)—a powerful network mod-
elling approach—to a rich dataset comprising seven online classes.
In the following sections, we rst construct a conceptual model of
social interaction in online discussions. We then introduce the REM
study and report its ndings. Finally, conceptual, pedagogical, and
methodological implications are discussed.

LAK ’20, March 23–27, 2020, Frankfurt, Germany Bodong Chen and Oleksandra Poquet
2 CONCEPTUAL FRAMEWORK
Conceptually, we posit that social interactions in online forums
need to be examined as events instead of states. Current research
is predominately focused on examining the state of “social rela-
tions” as static relationship networks. We argue for the need to
unpack the process in which a series of events engender a partic-
ular state. For an event, e.g., A posts a comment on a post made
by B, we need to recognize a “conceptual leap” often made by re-
searchers from such a socio-technical interaction (that involves
people, ideas, and technology) to an interpersonal interaction (i.e.,
A->B). This conceptual leap, often done unintentionally, has im-
portant consequences for our understanding of online discussions.
One signicant consequence is we lose sight of the nuanced context
in which an event occurs, as well as the unfolding events come
to shape future events. Therefore, we propose to investigate in-
teractions in online discussions as temporal networks that unfold
temporally through participation that focus on relational events in-
stead of states engendered by these events. To conceptually ground
this study, in this section we outline dierent factors that could
impact the formation of such temporal activity-driven networks in
online discussions.
2.1 Learner Heterogeneity in Posting Activity
Learners bring unique understandings, habits, and dispositions
into online discussions. Learners may set their own participation
goals based on personal interests and aspirations. They may have a
unique schedule of participation. Some would prefer more “listen-
ing” before posting ideas.
The analysis of relational events needs to nd ways to account
for learner heterogeneity. For example, to capture the intensity
of learners’ posting activity, one common approach is to count
the number of posts contributed by each learner. The timing and
spacing of each learner’s engagement can also be characterized
using log data [
23
]. In an early work, Macfadyen and Dawson [
16
]
showed a strong association between the number of discussion
replies posted and learner nal grade (r =.94, p<.0001). Joksimovic
et al. [
13
] note the number of posts as the most common measure
of social learning in MOOCs. Although studies have reported di-
verse evidence around the strength of relationship between learner
posting activity and grades, forum posts are commonly included in
modelling learner engagement and predicting completion [14].
2.2 Pedagogical Practices
Online discussions are not created equal. Research shows factors
such as discussion guidelines, deadlines, feedback, and instructor
presence aect the resulting discourse [
8
]. Types of discussion activ-
ities, such as discussing course readings, sharing new stories, peer
feedback, and building group knowledge, are all of educational value
and present in online and hybrid courses. The discussion guidelines
would naturally shape the nature and intensity of discussion. So
does the instructor’s decision to participate in the discussion [8].
Among these factors, time, in particular, shapes the learner’s
perception, experience, understanding of time online, and hereby
one’s online participation [
20
]. There is a delicate dance between the
instructor’s temporal expectations (e.g., participation deadlines) and
learner heterogeneity in individual posting that would contribute
to social structures in an online discussion space. Recent work
highlights the association between learners’ social positions and
their timing behaviors in online discussions; in particular, learners
organize themselves around the pedagogical structure of deadlines,
leading to varied conditions for social encountering and positioning
[4].
2.3 Social Dynamics
For a given pedagogical condition, there are a range of sociological
factors that could inuence social interaction. For example, while
instructor presence may set a general tone for class dialogues [
8
],
at a micro-level being the instructor poses superior pressure that
attracts student attention and interactions. Empirical studies, how-
ever, have arrived at mixed ndings, with some ndings showing
the instructor’s posts did not show higher probabilities of being
replied [
32
]. Peer pressure can have a positive impact on participa-
tion intent as well [
30
]. Even when the instructor sets participation
goals, active or “super” posters in a class could generate peer pres-
sure on a learner’s participation. Like in other social environments,
being popular (e.g., receiving more replies) helps attract even more
attention [
26
]; this eect of preferential attachment is observed in
scientic collaboration networks and social network sites [18, 19].
Reciprocity is also found to play a role in online discussions based
on quantitative and qualitative accounts [
1
,
6
]. These sociologi-
cal factors contribute to one’s social or relational capital in online
discussion environments, which in turn shapes forthcoming inter-
actions.
Figure 1: Conceptual model of social interaction in online
discussions
2.4 Technological Features
Besides individual, pedagogical, and social conditions, technolog-
ical congurations of online discussions are too often neglected
in research reports. As a matter of fact, technological features of
forums shape a learner’s interaction with technology and peers [
9
].
For example, when discussion posts are presented as a dynamic
hyperbolic tree instead of a discussion thread, students are found
purposefully selecting which discussion threads to read [
17
]. When

Socio-Temporal Dynamics in Peer Interaction Events LAK ’20, March 23–27, 2020, Frankfurt, Germany
online discussions happen in specialty software designed by re-
searchers, we need to be transparent about “pedagogical biases”
embodied in technological features [
25
]. When general purpose
tools such as Facebook and Twitter are used, we need to be clear
about how their technological features are mobilized to serve ed-
ucational purposes [
22
]. In addition, we need to interrogate ways
in which technological features and algorithmic decisions (e.g.,
sorting algorithms) implicitly shape interaction patterns in online
discussions [4].
Taken together, social participation and interaction in asynchro-
nous online discussions is a complex phenomenon shaped by vari-
ous factors. Figure 1 illustrates a conceptual model we developed to
guide our investigation of contributing factors. This model intends
to help us operationalize our study and is by no means comprehen-
sive. Cross-cutting factors may also exist and interact with factors
elaborated above.
3 METHODS
To examine social interaction in online discussions as networks un-
folding temporally, we adopted Relational Event Modelling (REM),
a framework that allows researchers to understand how networked
structures emerge from relational events and shape subsequent
interactions [
2
]. We applied REM to a secondary dataset from an
online discussion environment.
3.1 Description of Data
3.1.1 Context. The study was situated in seven online courses at a
large public university in the United States. Six courses were about
business management, information technologies, and ethics.
3.1.2 The Discussion Environment. All courses incorporated Yel-
lowdig, a discussion environment for educational use, as a primary
medium for peer interaction. Students in all classes participated
weekly on Yellowdig to discuss readings and share ideas. Dier-
ent courses have set dierent expectations of participation. For
example, one course required students to minimally contribute one
post in response to instructor prompts and comment on two posts
each week; another course asked students to generally share news
articles related to the course content and comment on each other
to earn up to 10 points each week.
3.1.3 Data Structure. The discussion environment Yellowdig re-
sembles social network sites like Facebook. Students could initiate
a post, known as a pin, that can be commented on or reacted to by
members of the class. Students could also mention each other in
a post or comment. Like most social network sites, Yellowdig was
also implemented with a news feed powered by a sorting algorithm,
which placed posts with the most recent activities on the top. As
such, technological features in Yellowdig are designed to promote
social interaction.
Interaction data from Yellowdig, including 6,354 discussion ac-
tions (e.g., making a pin, commenting on a pin, reacting to a pin)
from a total of 281 students, were the primary data source. Course
materials, including the syllabus and weekly announcements, were
gathered to contextualize the interpretation.
3.2 Data Analyses
3.2.1 Profiling Learners on Timing. Given that no demographic
information, performance, or discourse about the learners was
available for this study, modelling had very little information about
the learners. To this end, we adopted an approach similar to that
in Park et al. [
21
] where nite mixture models [
7
] were used to
group learners based on the total of their logs per each day of
the week. These latent classes captured how much platform use
learner engaged in on each given weekday. The latent classes, i.e.
those interpreted as time-consistent posters and last-minute posters,
were regressed to the volume of platform-use-logs, to help interpret
the clusters. Information about the timing of activity was used as
controls in the nal REM model – to improve the estimation of the
relationship between posting events and emergent social dynamics.
Clustered logs included diverse granular behaviors, namely: cre-
ating a pin (making an initial thematic post), visiting a pin, editing
a pin, editing a comment, liking a comment, ‘loving’ a pin, deleting
a pin, as well as creating and voting a pin.
We do not include any statistics related to the timing proles,
as this paper’s scope is limited to presenting the results of the
modelling, and timing proles were used only to capture similarity
in synchronous patterns of the learners reported in the nal REM
models for each course.
3.2.2 Relational Event Modelling. To address the major research
question about potential contributing factors, we applied a novel
network analysis approach named Relational Event Modeling (REM)
[
2
]. Detailed explanations of REM can be found elsewhere [
2
,
15
].
Formally, Butts and Marcum [
3
], expressed a relational event model
as
λaAtθ=(exp(θτu(s(a),r(a),c(a),Xa,At)) i f a ∈A(At)
0otherwise (1)
where
λ
represents the hazard of potential event aat time tgiven
history At; whereas a vector of real-valued parameters is
θ
; and
uis a vector of statistics, such as the sender, receiver, covariates,
prior event history (p.8).
Put simply, a relational event is dened as a “discrete event gen-
erated by a social actor and directed toward one or more targets”
[
2
, p. 159]. For example, Student A replies to B at time t1 is one
relational event a1, and B responds to A at a later time t2 is a newer
relational event that might have been triggered by a1. The cen-
tral goal of REM is to “understand how past interactions aect
the emergence of future interactions” [
2
, p. 533], based on a set of
derived statistics about (a) past relational events (e.g., frequency of
previous A
→
B events), (b) actor attributes (e.g., gender, roles), and
(c) exogenous contextual factors (e.g., friendship). Take the statistic
of reciprocity, for example. It indicates the likelihood of Student B
replying to A if A has recently commented on B. Statistics as such
capture both social and sequential information and are also referred
to as Sequential Structural Signatures (SSSs) of group dynamics
[
15
]. In this study, we trained multiple relational event models in-
corporating a range of SSSs, which could potentially shape peer
interaction in our research context. We tted these models follow-
ing a forward selection strategy and evaluated model adequacy

LAK ’20, March 23–27, 2020, Frankfurt, Germany Bodong Chen and Oleksandra Poquet
Table 1: Relational event modelling of interactive activity that comprised: created_pin, created_comment, like_pin,
liked_comment, love_pin, unliked_comment events.
Course 1 Course 2 Course 3 Course 4 Course 5 Course 6 Course 7
Node & Dyad Attributes
Sending (Time Consistent (TC)) 0.01 (0.1) 0.4 (0.1) *** 0.3 (0.1) 0.5 (0.2) *** 0.34 (0.1) * 0.2 (0.1)* 0.2 (0.1) .
Receiving TC 0.23 (0.1) * 0.02 (0.1) 0.4 (0.1) ** -0.42 (0.2) * -0.2 (0.2) 0.1 (0.1) 0.1 (0.1)
Homophily TC Controlled for
Sending Teacher 0.8 (0.2) *** 0.7 (0.2) *** 0.4 (0.2) 0.5 (0.2)** 1 (0.2) *** -0.8 (0.4)* 0.1 (0.3)
Receiving Teacher -2.1 (0.4) *** -2 (0.3) *** -0.8 (0.2) ** -0.5 (0.2) * -0.7 (0.3) * 0.3 (0.2) . 0.7 (0.2) ***
Structure
Degree (NTDegRec) 9.6 12.6 (0.8)*** 11.3 (1.4) *** 11.5 (1) *** 7.9 (1.3)*** 14 (1) *** 14 (1.2) ***
Reciprocity (RRecSnd) 0.47 (0.1)** 1.04 (0.1) *** 0.8 (0.1) *** 0.6 (0.2) ** 0.44 (0.2) . 0.3 (0.2) . 0.56 (0.2)***
Familiarity (RSndSnd) 1.06 (0.1)*** 2.42 (0.1) *** 2.2 (0.1) *** 2.3 (0.1)*** 1.6 (0.2) *** 2 (0.11) *** 1.29 (0.12)***
Closure (ISPSnd) 0.04 (0.01)*** 0.06 (0.02)*** -0.06 (0.1) ** 0.03 (0.01)* 0.03 (0.4) 0.01 (0.01) 0.03 (0.08)***
Immedaite Eects
Reciprocity (AB-BA) -0.33 (0.4) 2.3 (0.4)*** 1.8 (0.4) ** 1.5 (0.6)* 0.99 (0.8) 0.9 (1) 2 (0.5)***
Pay-it-forward (AB-BY) -2.4 (0.3)*** 0.9 (0.2) *** 0.83 (0.27) ** 0.1 (0.3) -0.1 (0.4) -0.1 (0.3) 0.56 (0.2)***
Activity bursts (AB-AY) 1.22 (0.1)*** 4.5 (0.1) *** 3.58 (0.1) *** 3.4 (0.1)*** 2.74 (0.2) *** 3.8 (0.1) *** 3.8 (0.1) ***
Popularity (AB-XB) -0.48 (0.1)*** 2.6 (0.1) *** 1.6 (0.2) *** 1.1 (0.2)*** 1 (0.2) *** 1.7 (0.1) *** 1.4 (0.2)***
AIC Null 6897 34130 6213 7616 3503 13053 12042
BIC Null 6897 34130 6213 7616 3503 13053 12042
AIC Final 5610 27201 4916 6286 3003 10756 9875
BIC Final 65610 27274 4969 6342 3051 10818 9936
Accuracy 0.72 0.68 0.78 0.69 0.67 0.68 0.67
based on the AIC (Akaike information criterion) and BIC (Bayesian
information criterion) scores [
2
]. A maximum likelihood estimate
(MLE) in a model provides an indication of the odds of a relational
event given the specied conditions.
4 RESULTS
4.1 Social and Temporal Dynamics
Outputs of the relational event models are reported in Table 1.
The rst set of hypotheses related to the propensity of individual-
level attributes to aect tie formation. Specically, we examined if
learners with time-consistent patterns of logging into the platform
to post were also more likely to send or receive ties (Hypotheses
1a and 1b). We observed that those learners who engaged with the
platform on the same weekdays with eort spread out across several
days (i.e. time-consistent) showed a positive tendency to send ties
to other learners. A signicantly positive propensity for these more
self-regulated learners to send ties was observed in three out of
seven models. These learners were also more likely to receive ties -
with positive and signicant coecients in two models. It is worth
noting that the sending and receiving dynamics for time-consistent
posters was not co-occurrent, i.e. we did not observe it in the same
courses. Generally, this suggests that timing of posting is a factor in
how networked structures of digital learning communication form.
Another set of hypothesis related to the role of the instructor
in how communication evolves (Hypotheses 1c and 1d). In four out
of seven models, the instructor was more likely to send ties. In six
out of seven models, the instructor was less likely to receive ties.
This suggests that instructors were present in the communication,
but their involvement was likely to facilitate inter-peer behaviors,
rather than instructor-learner interactions.
In terms of the structural elements, we modelled series of dier-
ent hypotheses.
H2a. Learner total activity aected future incoming ties. This hy-
pothesis was conrmed as positive and signicant in six out of
seven models, with large coecients for all courses. That is, the
more active the learner was, the more likely she was to receive
replies with the course progression.
H2b. Networks are described by temporal reciprocity, i.e. fraction
of ties from A to B, aected future sending ties of B to A. This hypoth-
esis was conrmed as positive and signicant in four out of seven
models. That is, the more replies a learner sent to the same person,
the higher the likelihood that this person would also be replying
with the progression of time.
H2c. Familiarity aects future tie formation, i.e. recency of sending
from A to B, aects tie formation from A to B. Recency of sending
ties was found to aect tie-sending in the future through positively
signicant coecient in all seven models. That is, learners were
more likely to reply to someone in the future, if they had engaged
with the person in a discussion previously.
H2d. Two learners who reply to the same third learner will form a
tie, i.e. closure hypothesis. Although closure was signicant in four
models, the coecients were low, suggesting that closure was not
captured well.
Finally, the third set of hypotheses examined diverse patterns of
sequential micro-level patterns. The eects related to immediacy in
learner interactions improved the model t signicantly across all
the courses. Immediate reciprocity, i.e. B would respond to A right
after A would send a tie to B, was conrmed through positive and
signicant coecients in four models (Hypothesis 3a). Immediate
two-path feature (A sending ties to B, B sending ties to C right
afterwards) that hypothesized that learners were likely to engage
if they received a reply, produced mixed results (Hypothesis 3b).

Socio-Temporal Dynamics in Peer Interaction Events LAK ’20, March 23–27, 2020, Frankfurt, Germany
Overall the coecients were low, and positive or negative across
the models. Perhaps one of the strongest immediacy parameters
was that of activity bursts (Hypothesis 3c). Sending ties consequently
to dierent learners (A to B, A to C) in a two-star pattern showed
large coecients that were positive and signicant in all seven
models. This reects the nature of posting behaviour as a part of
assignment - when learners respond to several learners as they log
into the platform. Finally, popularity, or more precisely, sequential
incoming replies, modeled as a temporal in-degree 2-star (A to B, C
to B) showed positive and signicant coecients across all models,
except one (Hypothesis 3d).
4.2 Goodness of Fit
Although all the nal REM models (tted iteratively to capture
degree, reciprocity, closure, as well as exogenous factors, and imme-
diacy patterns) showed improvement in AIC from the null model
(reported in the Results, Table 1), the accuracy of prediction var-
ied. We examined what events (posting activity) the models were
failing to predict. From the plots of surprising events (i.e. events
that the models did not predict well), it appeared that closure was
not explained by the closure patterns included in the model. In
other words, discussion online was resulting in patterns of triads
that were not explained by any of the parameters included in the
models. Although we controlled for possible homophily through
attributes of learner activity and learner timing, the models still
did not capture clustering well. One reason for this could be that
these models exclusively used log data and timestamps from the
discussion forums. We suggest that the closure in the network is
subject to the text posted by learners that mediates these peer in-
teractions. It is also plausible that the closure resulted from visiting
someone’s pin that was found attractive. Perhaps, interviews with
students and collection of self-reports around their reactions to
posted discourse can help shed light on the potential dynamics.
5 DISCUSSION
The study is framed from the perspective that interactions in digital
learning settings would benet from focusing on relational events
rather than relational states. To demonstrate the possibility, we
applied Relational Event Modeling (REM) to a dataset from seven
online classes. In the modelling, we included a learner attribute of
temporal participation, alongside socio-temporal factors such as
preferential attachment and reciprocity formed through timing or
sequences of peer interaction. We found that temporality adds a new
dimension to the analyses of peer interactions. Social dynamics and
temporal dynamics in these networks are intertwined as showed
by improvement in the models when adding immediacy eects.
Activity is heavily driving network formation, whereas social eects
of reciprocity and closure are rather low.
Approaching digital peer interactions as relational events high-
lights that peer event networks are formed by learner activity.
Teacher presence was observed as signicant across most courses,
but it appears not to impact learner reciprocal activity towards the
instructor. Instead, tie formation between the learners seems to
result from rather random processes, such as familiarity based on
recency and co-occurrence. We have not observed the eect from
homophily based on timing controls.
REMs also reveal that temporal behaviors at the dyadic level may
emulate structures similar to social dynamics (i.e. transitivity, two-
paths, or two-stars). Yet, failure of the model to predict clustering
in the network suggests that these patterns only look like social
dynamics, without necessarily representing it.
Empirical ndings from this study are of scholarly and practical
signicance. Compared to earlier studies applying REM in educa-
tional contexts, we theorised dierent processes of network forma-
tion in online discussions, applied REM to a unique context, and
combined temporal and social dimensions of network formation. To
better understand underlying social dynamics in such text-mediated
peer interaction events, we suggest that future work should focus
on the role of knowledge construction behaviors in relation to
triadic closure patterns and homophily. It may be plausible that
learners exchange interactions based on demographic variables
underpinning their identities. However, it is more likely that the
complementarity of learner communication strategies drives the
interpersonal aspects of their interaction. Future work can also
explore comparing and combining REM with discourse analyses.
This study contributes fresh insights into social interaction in
online discussions, calls for attention to micro-level temporal pat-
terns, and motivates future work to scaold learner participation in
similar contexts. Results suggest that including information about
temporality of peer interaction events is relevant. However, further
information is required to better understand what mediates these
interaction events.
REFERENCES
[1]
Reuven Aviv and Gilad Ravid. 2005. Reciprocity analysis of online learning
networks. Journal of Asynchronous Learning Networks 9, 4 (2005), 3–13.
[2]
Carter T Butts. 2008. A Relational Event Framework for Social Action. Sociological
Methodology 38, 1 (2008), 155–200.
[3]
Carter T Butts and Christopher Steven Marcum. 2017. A relational event approach
to modeling behavioral dynamics. In Group Processes. Springer, 51–92.
[4]
Bodong Chen and Tianhui Huang. 2019. It is about timing: Network prestige in
asynchronous online discussions. Journal of Computer Assisted Learning (2019).
[5]
Bodong Chen, Marlene Scardamalia, and Carl Bereiter. 2015. Advancing
knowledge-building discourse through judgments of promising ideas. Interna-
tional Journal of Computer-Supported Collaborative Learning 10, 4 (2015), 345–366.
[6]
Wing Sum Cheung, Khe Foon Hew, and Connie Siew Ling Ng. 2008. Toward an
understanding of why students contribute in asynchronous online discussions.
Journal of Educational Computing Research 38, 1 (2008), 29–50.
[7]
Partha Deb and Pravin K Trivedi. 2013. Finite mixture for panels with xed
eects. Journal of Econometric Methods 2, 1 (2013), 35–51.
[8]
Vanessa Paz Dennen*. 2005. From message posting to learning dialogues: Factors
aecting learner participation in asynchronous discussion. Distance Education
26, 1 (2005), 127–148.
[9]
Mark Guzdial and Jennifer Turns. 2000. Eective discussion through a computer-
mediated anchored forum. The journal of the learning sciences 9, 4 (2000), 437–469.
[10]
Jim Hewitt and Clare Brett. 2011. Engaging learners in the identication of
key ideas in complex online discussions.. In Connecting Computer-Supported
Collaborative Learning to Policy and Practice (CSCL2011 Conference Proceedings).
960–961.
[11]
I Jo, Yeonjeong Park, and H Lee. 2017. Three interaction patterns on asynchronous
online discussion behaviours: A methodological comparison. Journal of Computer
Assisted Learning 33, 2 (2017), 106–122.
[12]
Srećko Joksimović, Dragan Gašević, Vitomir Kovanović, Bernhard E Riecke, and
Marek Hatala. 2015. Social presence in online discussions as a process predictor
of academic performance. Journal of Computer Assisted Learning 31, 6 (2015),
638–654.
[13]
Srećko Joksimović, Oleksandra Poquet, Vitomir Kovanović, Nia Dowell, Caitlin
Mills, Dragan Gašević, Shane Dawson, Arthur C Graesser, and Christopher
Brooks. 2018. How do we model learning at scale? A systematic review of
research on MOOCs. Review of Educational Research 88, 1 (2018), 43–86.
[14]
René F Kizilcec and Emily Schneider. 2015. Motivation as a lens to understand
online learners: Towarddata-driven design with the OLEI scale. ACMTransactions
on Computer-Human Interaction (TOCHI) 22, 2 (2015), 6.

LAK ’20, March 23–27, 2020, Frankfurt, Germany Bodong Chen and Oleksandra Poquet
[15]
Contractor Noshir Leenders, Roger and Leslie DeChurch. 2016. Once upon a
time: Understanding team processes as relational event networks. Organizational
Psychology Review 6, 1 (2016), 92–115.
[16]
Leah P Macfadyen and Shane Dawson. 2012. Numbers are not enough. Why
e-learning analytics failed to inform an institutional strategic plan. Journal of
Educational Technology & Society 15, 3 (2012), 149–163.
[17]
Farshid Marbouti and Alyssa Friend Wise. 2016. Starburst: a new graphical
interface to support purposeful attention to others’ posts in online discussions.
Educational Technology Research and Development 64, 1 (2016), 87–113.
[18]
Alan Mislove, Hema Swetha Koppula, Krishna P Gummadi, Peter Druschel, and
Bobby Bhattacharjee. 2008. Growth of the ickr social network. In Proceedings of
the rst workshop on Online social networks. ACM, 25–30.
[19]
Mark EJ Newman. 2001. Clustering and preferential attachment in growing
networks. Physical review E 64, 2 (2001), 025102.
[20]
Murat Oztok, Lesley Wilton, Kyungmee Lee, Daniel Zingaro, Kim Mackinnon,
Alexandra Makos, Krystle Phirangee, Clare Brett, and Jim Hewitt. 2014. Polysyn-
chronous: dialogic construction of time in online learning. E-learning and digital
media 11, 2 (2014), 154–161.
[21]
Jihyun Park, Renzhe Yu, Fernando Rodriguez, Rachel Baker, Padhraic Smyth, and
Mark Warschauer. 2018. Understanding Student Procrastination via Mixture
Models. International Educational Data Mining Society (2018).
[22]
Peter Ractham, Laddawan Kaewkitipong, and Daniel Firpo. 2012. The use of Face-
book in an introductory MIS course: Social constructivist learning environment.
Decision Sciences Journal of Innovative Education 10, 2 (2012), 165–188.
[23]
Jeremy Riel, Kimberly A Lawless, and Scott W Brown. 2018. Timing Matters:
Approaches for Measuring and Visualizing Behaviours of Timing and Spacing of
Work in Self-Paced Online Teacher Professional Development Courses. Journal
of Learning Analytics 5, 1 (2018), 25–40.
[24]
Christina Sadowski, Mika Pediaditis, and Robert Townsend. 2017. University
students’ perceptions of social networking sites (SNSs) in their educational expe-
riences at a regional Australian university. Australasian Journal of Educational
Technology 33, 5 (2017).
[25]
Marlene Scardamalia and Carl Bereiter. 2008. Pedagogical biases in educational
technologies. Educational Technology (2008), 3–11.
[26]
Luis Vaquero and Manuel Cebrian. 2013. The rich club phenomenon in the
classroom. Scientic reports 3 (2013), 1174.
[27]
C-H Wang. 2005. Questioning skills facilitate online synchronous discussions.
Journal of Computer Assisted Learning 21, 4 (2005), 303–313.
[28]
Alyssa Wise and Ming Ming. Chiu. 2011. Knowledge construction patterns in
online conversation: A statistical discourse analysis of a role-based discussion
forum.. In Connecting Computer-Supported Collaborative Learning to Policy and
Practice (CSCL2011 Conference Proceedings). 64–71.
[29]
Samantha Won, Michael Evans, and Lixiao Huang. 2017. Engagement and knowl-
edge building in an afterschool STEM Club: analyzing youth and facilitator
posting behavior on a social networking site. Learning, Media and Technology 42,
3 (2017), 331–356.
[30]
Xue Yang, Yan Li, Chuan-Hoo Tan, and Hock-Hai Teo. 2007. Students’ partic-
ipation intention in an online discussion forum: Why is computer-mediated
interaction attractive? Information & Management 44, 5 (2007), 456–466.
[31]
Binbin Zheng and Mark Warschauer. 2015. Participation, interaction, and aca-
demic achievement in an online discussion environment. Computers & Education
84 (2015), 78–89.
[32]
Daniel Zingaro and Murat Oztok. 2012. Interaction in an asynchronous online
course: A synthesis of quantitative predictors. Journal of Asynchronous Learning
Networks 16, 4 (2012), 71–82.