Computers in Human Behavior 64 (2016) 782e792

Contents lists available at ScienceDirect

Computers in Human Behavior

journal homepage: www.elsevier.com/locate/comphumbeh

Full length article

Towards computational discourse analysis: A methodology for mining

Twitter backchanneling conversations
Carlo Lipizzi b, Dante Gama Dessavre b, Luca Iandoli a, b, *,
Jose Emmanuel Ramirez Marquez b
a
Dept. of Industrial Engineering, University of Naples Federico II, Piazzale Tecchio 80, 80125 Napoli, NA, Italy
b
School of Systems and Enterprises, Stevens Institute of Technology, Castel Point on Hudson 1, 07030 Hoboken, NJ, USA

a r t i c l e i n f o a b s t r a c t

Article history: In this paper we present a methodology to analyze and visualize streams of Social Media messages and
Received 19 March 2016 apply it to a case in which Twitter is used as a backchannel, i.e. as a communication medium through
Received in revised form which participants follow an event in the real world as it unfolds. Unlike other methods based on social
3 July 2016
networks or theories of information diffusion, we do not assume proximity or a pre-existing social
Accepted 22 July 2016
structure to model content generation and diffusion by distributed users; instead we refer to concepts
Available online 8 August 2016
and theories from discourse psychology and conversational analysis to track online interaction and
discover how people collectively make sense of novel events through micro-blogging. In particular, the
Keywords:
Online discourse analysis
proposed methodology extracts concept maps from twitter streams and uses a mix of sentiment and
Conversational analysis topological metrics computed over the extracted concept maps to build visual devices and display the
Social media mining conversational flow represented as a trajectory through time of automatically extracted topics. We
Social representations evaluated the proposed method through data collected from the analysis of Twitter users’ reactions to
Big data the March 2015 Apple Keynote during which the company announced the official launch of several new
Online communities products.
Online conversation © 2016 Elsevier Ltd. All rights reserved.
Computer-mediated communication
Collective intelligence
Semantic analysis
Sentiment analysis
New product launch

1. Introduction breadth in which a large number of users generates and shares

The large-scale adoption of Social Media (SM) is one of the most
relevant technological and social trends in the history of the
Internet. According to the Pew Research Center (Duggan, 2015),
about three quarters of adult Internet users in the US spend
considerable time on Social Network sites such as Facebook,
Twitter, Instagram, Pininterest, and LinkedIn, with 65% of online
adults using social media sites (Perrin, 2015), with a tenfold in-
crease in ten years (this percentage goes up to 70% and 59%,
respectively, for the adults who use daily Facebook and Instagram).
SM are thus creating interaction spaces of unprecedented size and
* Corresponding author. Dept. of Industrial Engineering, University of Naples
Federico II, Piazzale Tecchio 80, 80125 Napoli, NA, Italy.
E-mail addresses: carlo.lipizzi@stevens.edu (C. Lipizzi), ddessavre@stevens.edu
(D.G. Dessavre), luca.iandoli@unina.it, luca.iandoli@stevens.edu (L. Iandoli),
jmarquez@stevens.edu (J.E. Ramirez Marquez).
contents through hyper-connected social networks while trying to
accomplish various types of tasks such as voicing their opinion,
providing or asking for help, sharing information, contributing to a
cause, reaching out to friends and acquaintances, or applying for
memberships to a community.
Much of this interaction happens through online conversations
that can be tracked and mined to extract online analytics for
different applications. A broad practical question for media analysts
is the availability of reliable tools to summarize large conversa-
tional flows into effective representations to answer a number of
“what” and “who” questions such as: what the users are talking
about? What matters most to them? What is trending right now?
Who is talking about what? Who is talking to whom? Etc. The
answers to such questions are of immediate use in multiple ap-
plications in fields as diverse as marketing (Hoffman & Fodor, 2010,
pp. 1e11; Smith, Rainie, Shneiderman, & Himelboim, 2014), politics
(Bartlett, Froio, Littler, & McDonnell, 2013; Parker, 2014), or

http://dx.doi.org/10.1016/j.chb.2016.07.030
0747-5632/© 2016 Elsevier Ltd. All rights reserved.
 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
national security (Brachman, 2014; Klausen, 2015). Moreover,
thanks to the possibility to track online interaction in an objective
fashion, online conversations offer plenty of interesting empirical
materials for the analysis and modelization of learning in virtual
social networks, although predominantly in research on use of SM
in education (Greenhow & Greenhow, 2011; Yu, Tian, Vogel, &
Kwok, 2010).
In this paper our approach to answer “what” and “who” ques-
tions is based on tracking and visualizing the process through
which large-scale interaction favors the emergence of shared
meaning through a process of participatory sense-making (Jaegher
& Paolo, 2007). As we show later in the paper (Section 2), the
analysis of this process has received limited attention in the liter-
ature on online conversations, which instead could benefit from the
application of computational methods based on natural language
processing, automated semantic analysis, and social network
analysis as well as from theoretical and methodological insights
from studies in conversational analysis and discourse psychology
(Section 3).
Through the development of a methodology that combines
ideas and tools from conversation analysis, online analytics, and
data visualization (Section 4), we provide a methodology and a tool
to better handle the trade-off between in depth, small-scale,
qualitative analysis and processing of large amount of data gener-
ated in conversations. We then show that SM-enabled conversa-
tions can support sense-making about new events by favoring the
accumulation of common ground generated by multiple partici-
pants when following a same event in real time (Clark & Brennan,
1991) (Section 5).
Among the several available SM, we focus on Twitter for a
number of reasons. Twitter is popular and widespread among
Internet users (Duggan, 2015) and the access to its data is relatively
easy and low cost with most users making their content public.
Tweets lack the complexity and richness of more structured mes-
sages, such as those that can be found in online forums, thanks to
the well-known 140 characters constraint. Finally, Twitter is
increasingly used as a backchannel to follow up with events
unfolding in the real world (Dork, Gruen, Williamson, &
Carpendale, 2010). Twitter backchannel feeds are also frequently
used to support virtual participation to a real event (McNely, 2009)
by providing participants with a parallel communication channel to
exchange comments about what is going on.
Backchanneling represents an interesting case for the empirical
analysis of SM because the short duration of the events being fol-
lowed helps to create a strong focus in the discussion, unlike the
more unstructured and chaotic feature of uneventful micro-
blogging. More importantly, backchanneling is akin to virtual
participation to a real world event, which participants are trying to
make sense of and to which they are attracted for a variety of in-
dividual motivations. We suggest that this sense-making process is
easier to observe in those cases in which the participants are
generally ignorant about the ways the event is going to unfold,
although they may have some pre-existing knowledge and expec-
tations about its development and the subject in general. This is the
case for events such as a TV political debate (Shamma, Kennedy, &
Churchill, 2009), a new episode of a TV show (Harrington,
Highfield, & Bruns, 2012), and the use of SM to coordinate in
emergency situation (Hughes & Palen, 2009; Mills, Chen, Lee, &
Raghav Rao, 2009).
In the next section we show that the dominant approaches to
SM mining are biased towards the analysis of user-generated con-
tent rather than of the process through which the same content is
generated. Our challenge is instead to dig deeper into Twitter
streams to observe the emergence of shared meaning in online,
large-scale conversations by using a quantitative methodology
783
inspired to ideas and theories from studies in discourse analysis.
The proposed methodology is applied to extract social and concept
maps from micro-blogs streams and to compute a set of conver-
sational analytics for the design of visual devices able to display the
conversational flows.
Since the maps are generated as the conversation unfolds, our
approach helps to analyze how the structure of these graphs
changes in time and reflect not only what the participants are
talking about but also how they are talking about the event.
Structural analysis is then used to visualize the conversational flow
and identify emergent conversational patterns and the process that
generates them. The combination of social, semantic, and more
traditional traffic metrics helps to observe conversational interac-
tion and its outcome in a dynamic fashion, possibly offering in-
sights on the role of social cognition on meaning generation and
transformation (Jaegher & Paolo, 2007) and empirical grounding to
better justify the use collaborative/social technologies to support
informal and continuous learning (Siemens, 2005).
We report empirical findings obtained from one case study
related to the launch of new products, the Apple Keynote presen-
tation that took place in March 2015, a 100 min event during which
the company launched the new MacBook and the much-awaited
AppleWatch. We selected this event as an example of a social
happening that is highly focused, that is object of wide attention
and ample coverage from other media, and that is scheduled
regularly by Apple twice per year. The application to a marketing
case is also used to offer an example of practical applications, but
the proposed method can be applied to other events that exhibit
similar characteristics in terms of focus, popularity, and
expectations.
2. Current approaches to twitter streams analysis
2.1. Overview
The availability of reliable and efficient methods for the collec-
tion and analysis of SM data is an increasing concern for companies
and policy makers (Hoffman & Fodor, 2010, pp. 1e11; Smith et al.,
2014). In particular, without the ability to listen to and measure
SM flows, media analysts miss a valuable opportunity to exploit the
abundant and informative data that is incessantly generated by
millions of online users and transform this information into
actionable analytics to support more responsive and data-driven
decision-making (Culnan, McHugh, & Zubillaga, 2010).
These needs bring new online analytic challenges due to the
growing variety of online media to be monitored and the conse-
quent production of massive quantity of data to be analyzed. For
instance, different types of SM enable different types of conversa-
tional tasks and allow users to assume multiple online identities,
especially among younger users (Bolton, Aksoy, van Riel, &
Kandampully, 2013; Williams, Crittenden, Keo, & McCarty, 2012).
Facebook, for instance, can be used for sharing personal contacts,
news and experiences with friends, family and other acquaintances.
Twitter is born with the idea of giving people a simpler tool to
answer the question: what are you doing now? Lately, it has
evolved into a public space in which people broadcast their opin-
ions or whatever they think is worthwhile of public attention. SM
conversations are also very different from online forums, which are
often directed towards a common topic and support longer lasting
conversations that are attended by a smaller number of
participants.
Consequently, alternative ways to visualize these new types of
social streams have been proposed recently while methods devel-
oped to analyze more “traditional” types of online conversations
such as in online forum (Melville, Sindhwani, & Lawrence, 2009) or
784 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
email exchanges (Vie !gas, Golder, & Donath, 2006) are not appro-
priate to the always-on, large scale, fragmented conversations that
take place on the new media.
In this paper specifically we focus on an application of Twitter
that is achieving increasing popularity known as backchanneling
(Dork et al., 2010). Backchanneling is the practice of using com-
puters to sustain a real-time, online conversation alongside a pri-
mary event that is unfolding offline or on another medium, such as
TV shows and conferences (Wikipedia definition). Twitter back-
channeling data provide a more focused opportunity for SM min-
ing. Examples include monitoring TV audience reactions (Highfield
& Bruns, 2013; Nielsen, 2013; Wakamiya, Lee, & Sumiya, 2011),
understanding ad predicting traders’ reactions in the stock market
(Evangelopoulos, Magro, & Sidorova, 2012), or measuring the
diffusion of ideas in online discussions (Tsur & Rappoport, 2012). In
all of these cases the objective is to extract meaningful and repre-
sentative feedback from online conversations and exploit this
feedback to help decision makers to identify effective courses of
actions.
Because the amount of data to track and analyze in back-
channeling conversations can easily become huge, the empirical
analysis of user-generated content typically requires the adoption
of quantitative and automated data mining tools. The most
commonly used approaches in social web mining are based on
theories of Information Diffusion in Networks. Another important
category of methods is based on computer-based semantic analysis
and includes applications such as sentiment and reputation anal-
ysis or text mining of the content published on SM site through
natural language processing algorithms.
Diffusion theory provides conceptual and methodological tools
to assess the rate of information circulation throughout a social
network and has been applied to off-line as well as to online studies
such as the adoption of new ideas or practices in a community
(Rogers, 1995), the spreading of computer viruses (Vespignani,
2005) or the implementation of viral marketing in online net-
works (Leskovec, Adamic, & Huberman, 2007). The unit of analysis
is the structural organization of connections (“topology”) between
users in a social network and the ways through which it favors or
hinders the diffusion of a message. Social graphs can be analyzed
through social network analysis models (SNA) aimed at repre-
senting structural properties of networks such as centrality, links
distribution (Barab! asi, Albert, & Jeong, 2000), embeddedness (Uzzi,
1997), ties strength (Granovetter, 1973), and structural holes (Burt,
2004).
Other and more recent computer-enabled methods are based on
semantic analysis and natural language processing techniques for
the automated extraction of meaning and sentiment from the
messages that are exchanged in a network (Brown, 2012; He, Zha, &
Li, 2013; Saif, He, & Alani, 2012). Statistical methods to extract in-
sights from the text are also widely used. For example, frequency
weighting (such as term frequency - inverse document frequency,
td-idf) is a popular technique used to extract important terms out of
a corpus of documents according to their relevance (Manning,
Raghavan, & Shutze, 2009, pp. 1e569). Another commonly used
method to analyze text is co-occurrence. According to this method,
two or more words are considered semantically related when they
frequently appear within a given range of proximity in the text
(degree of separation). Using co-occurrence, it is possible to create a
network of words, with nodes being the words and edges being
their co-occurrences. Creating topological representations from
words in text based on word co-occurrence has been applied to
extract information on the structure of the text in online chats
(Introne & Drescher, 2013).
The design of effective tools to represent SM content is a priority
for the development of commercial products, generally known as
“listening platforms” (Gartner, 2013, pp. 1e98), aimed at displaying
social data in visual dashboards to support analysts and decision
makers (Cui et al., 2011; Dork et al., 2010; Pousman, Stasko, &
Mateas, 2007; Sack, 2000). Companies adopt listening platforms
to exploit social intelligence and turn SM data into actionable
marketing and business knowledge. SM metrics may be based on
raw indicators, such as number of comments posted, friends
reached, and the like; however, the creation of added value metrics
usually requires meaningful and smart aggregation of raw counters,
such as sentiment, buzz, rating, influence, exposure, and virality
into compound indicators. Opinion mining and sentiment analysis
are part of the standard features of many commercial platforms for
monitoring SM, such as Metavana, Sysomos, or Crimson Hexagon.
Sentiment analysis is applied to automatically identify and assess
value judgments expressed by online users with respect to a spe-
cific object (e.g., a brand) (Chamlertwat, Bhattarakosol, &
Rungkasiri, 2012).
2.2. Limitations of current methods
None of the methods and approaches outlined above assumes
conversational interaction as the primary unit of analysis but they
rather focus on the output generated by the users or on the social
infrastructure that enables information exchange.
Information diffusion and Viral models are content-neutral i.e.,
they typically assume that meaning is not problematic and that it is
preserved when information is passed along social links while little
attention is paid to the recursive and cyclic nature of the interaction
taking place between users. This view of information as a neutral
token to be passed between an active sender and a passive receiver
is due to the influential channel metaphor proposed by Claude
Shannon (Sloane & Wyner, 1948, pp. 5e83). One consequence is
that SNA, or viral models for that matter, work better when the
semantic alteration of content produced by users is not problem-
atic, e.g. when diffusion is about transferring packaged information
(e.g. a computer virus) or very short pieces of content, e.g. Internet
memes (Bauckhage, 2011).
Approaches based on semantic analyses are characterized by a
similar bias: the focus is on the automated extraction of meaning
from the messages propagating in a network, but little attention is
paid to the social dynamics through which messages are created
and modified while diffusing in the network. Semantic analysis
methods also have technical limitations when applied to SM min-
ing because most of these methodologies have been developed for
more coherent and traditional texts rather than for analyzing
conversations. While td-idf and its variations are used in several
applications such as in search engines (Croft, Metzler, & Strohman,
2010), their applicability to SM conversation has been questioned
(Hu & Liu, 2012; Qian, Zhou, Zhang, & Zhao, 2013). Similarly, co-
occurrence cannot be easily applied to dispersed online conversa-
tions e such as those happening on Twitter - because the text,
typically a sparse collection of short messages produced by multi-
ple authors in a dynamic way, often lacks the coherence of a written
document or of conversations directed at discussing specific topic
(Naveed, Gottron, Kunegis, & Alhadi, 2011). Because of these limi-
tations, sentiment/opinion analysis is often chosen as the main
option. Sentiment analysis, however, is just concerned with the
polarity of messages and does not provide any means to analyze
their content and discursive function.
Table 1 summarizes the most common SM mining approaches in
terms of key assumption, dominant metaphor and modelization
approach.
Diffusion theories, including information exchange in social
networks and viral models, assume that social connections preexist
to the actual exchanges. The structure of the relational or proximity
 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
Table 1
Dominant approaches to SM analysis.
Approach Key assumption
Information exchange in Information percolates throughout social links
social networks
Viral models Information is diffused through contagion mechanism that requires
physical, virtual or cognitive proximity
Semantic and Sentiment Meaning can be detected from messages via lexical, statistical, or semantic Information
Analysis analysis
Listening Platforms Data visualization supports decision Undisclosed makers' insights
network is thus given and any feedback between the content being
exchanged and the formation of the network is typically
disregarded.
Semantic and Sentiment analyses assumes that conversational
interaction can produce order and structure, for instance in terms of
positive or negative collective feeling; however, neither approach
investigates how these structures emerge from or are influenced by
interaction.
Finally, Listening Platforms rely heavily on data visualization
devices aimed at facilitating interpretation by analysts and decision
makers, but these representations are not typically driven by the-
ory; moreover, since visual dashboard are developed as commercial
tools, the assumptions and the algorithms behind the tools are
often undisclosed and there is no way to assess them in terms of
their theoretical grounding.
In conclusion, most of the existing tools and approaches are not
aimed at detecting conversational patterns in SM streams. In the
next section we will argue instead that the identification of
conversational patterns can help to assess not only how people feel
about a new product (sentiment) but to develop a better under-
standing of what people say and how people talk to each other.
Following some key tenets of Discursive Psychology, we explore the
constructive and pragmatic nature of online conversations in SM
backchanneling and try to incorporate some of the ideas and
principle used in conversational analysis in a computational
methodology for the analysis of online social streams in Twitter.
3. A discursive psychology approach to online conversational
analysis of Twitter streams
In this paper we follow a theoretical perspective based on
Discourse Psychology (DP) (Potter & Edwards, 1999), Social Rep-
resentation theory (Moscovici, 2001), and conversational analysis
(Clark & Brennan, 1991) to model micro-blog streams generated in
backchanneling applications as loosely-coupled conversations. DP
questions the cognitivist view according to which language is a
means to access what is in people mind and assumes instead a
pragmatic view in which utterances are linguistic acts people
perform to accomplish specific tasks (Clark, 1996). A fundamental
meta-task is the need to understand each other in conversations
through the accumulation of shared knowledge. According to
Potter and Edwards (1999) in DP shared knowledge representations
are discursive objects that people construct in talk and texts and
the objective of conversational analysis is to investigate the way
these representations are constructed and used to perform action
(e.g. assigning blame, eliciting invitations, and so on).
The kinds of questions that discursive psychologists might ask
regard how particular descriptions of a concept are used to do
particular things. In the cases analyzed in this paper, the official
launch of a new, untested product around which there is a lot of
expectation, but whose features, uses and even market positioning
are unclear, is a case in which potential customers look for feedback
785
Dominant Modeling
metaphor
Communication Communication networks modeled
channel through sociographs
Contagion Proximity Networks, infection likelihood
and threshold
Statistical techniques
processing Automated classifiers of the polarity of
Social mood individual judgments
Traffic, flow Data visualization techniques
from their peers to accomplish a variety of tasks such as: a) un-
derstand what's new and exciting with the new product; b) who is
buying in and why; c) find reasons, motives, and evidence to justify
purchasing intentions.
Following Clark and Brennan (1991) we assume that an online
conversation is a form of joint action requiring participants to co-
ordinate on content and on process. Coordination on content re-
quires contributors to achieve mutual understanding and build
common ground by asking for and providing evidence to confirm
they understand each other and by adding validated new infor-
mation to their stock of mutual knowledge. Feedback comes
through verbal requests (“what do you mean by … ?”) or nonverbal
acts like nodding, frowning, pointing, humming, etc. Coordination
on process is based on the reply structure; turn-taking serves the
purpose of achieving consensus that a new piece of information
adds up to the stock of mutual knowledge.
The narrow focus on a specific event that characterizes back-
channeling streams helps participants to achieve coordination on
content, while the use of other Twitter functions supports a mini-
mal level of reply structure. These functions include “mentions”
(mentioning the Twitter id of another user in the tweet, @user_id),
“retweets” (reposting a tweet initially posted by another users,
RT:<tweet〉), and “replies” (reply directly to another user, Reply
to@user_id).
DP and conversational analysis offer a set of theoretical con-
structs and mechanisms to model conversational content gener-
ated on micro-blogging platforms.
First, micro-blogs conversations may lead to the construction of
shared representations; however, since process coordination is
constrained by the limited affordances that Twitter offers in terms
of reply structure, we expect these representations to be more
fleeting than they would be in a real, longer term discourse. Sec-
ond, the concept of adjacency pairs (Clark & Schaefer, 1989) can be
exploited to map the content of tweets over a collective concept
map that is updated as the conversation proceeds. In ordinary
conversations adjacency pairs are two ordered utterances pro-
duced by speakers, as in a question/answer pair (e.g.: A: “Where is
Connie?”, B: “At the store”). A conversation can then be repre-
sented as a stream of adjacency pairs uttered within a turn-taking
structure.
In a Twitter conversation adjacency pairs are infrequent because
most of the time participants do not reply to each other (Sysomos,
2010) and, even when this happens, the conversational flow is
dispersed in a linear, time-based visualization of posts. It is then
more likely that users follow the stream by visualizing local por-
tions of it by proximity to their own or others’ contribution of in-
terest. This local focus helps to sustain a conversational scope that
is situated in a specific time and screen space. In our methodology
we model this aspect by reframing adjacency in terms of socio-
semantic proximity: two contributions are related because they
overlap spatially; two users are related because they are talking
about a same entity. As we show in the next section, this proximity
786 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
perspective is used to extract the content generated by users in a
twitter stream and reassemble it into a concept map that is updated
as the conversation proceeds.
A concept map is a network of keywords that can be analyzed
through topological and semantic metrics. For instance, it may
exhibit a certain arrangement of nodes and links (topology) and
some degree of internal lexical diversity. Keywords may be orga-
nized in emerging topical clusters. Finally, a number of network
metrics can be used to measure the structural properties of the
network and how they evolve in time. In the following section we
show how we implemented these ideas through a methodology
that is able to dynamically and automatically extract and visualize
concept maps from twitter streams; we then present semantic
metrics and data visualization formats to monitor the discursive
interaction taking place in Twitter.
4. Methodology
In this section we describe a methodology and its related soft-
ware tools we have developed to extract metrics and contents from
micro-blog streams. We then use metrics and contents as inputs for
visual tools designed to display how a backchanneling conversation
unfolds in time. In the following we provide a step-by-step
description of the procedure we developed to build such
GW ¼ G & GT ¼ ðW; Aw Þ where Aw ¼ fðp; qÞ; d x2U : ðp; xÞ2A and ðq; xÞ2Ag
visualization.
The core step of the method is the extraction of the semantic
and social structure. In particular in step 2 we operationalize the
idea of logic adjacency pairs and, in step 3, the concept of reply
structure, as described in the previous section. We then represent
common ground accumulation through dynamic concepts maps
that are built and modified as the conversation proceeds and from
which our algorithm is capable of extracting clusters of keywords
corresponding to emerging topics. A metric of semantic continuity
is then proposed in step 4 to detect and display changes in the
common ground accumulation process as signaled by semantic
switches occurring during the conversation.
4.1. Step 1: Data collection and pre-processing
In this step, an event around which a backchannel conversation
is likely to be started by interested users must be identified. Back-
channeling events have to be popular enough to attract the atten-
tion of a reasonably large number of users for a short time.
Examples of such events can be the launch of a new product by a
famous company, a live sport, or a TV show.
The tweets are downloaded via Twitter search Application
Program Interface (API), through a Python scripts, using keywords
related to the event. In addition to their 140 characters text, tweets
contain a large amount of meta-information, which includes items
such as user id, timestamp to location, language, and additional
information on the users. For our analysis we extract from each
tweet only a triple < u; W; t > where u2U is a user sending the
tweet, W is a vector of words wp 2W composing the text of the
tweet, and t is the time the tweet has been posted. The word vector
is obtained as follows:
! Tokenization: the raw text in each tweet is mapped onto a word
vector;
! Stop-wording: we filter the words from the previous step using
a customizable list of words we want to remove from the con-
versation (e.g. stop words, search keyword, etc.)
! Lemmatization: inflected forms of a same word are grouped
together so they can be reduced to a single base lemma (e.g.
work is the base lemma for worker, worked, working etc.).
! Bi-gramming: we extract most frequent bigrams from the
stream of tweets and process them as a single word (e.g.
Apple_watch).
The dataset is partitioned in time windows that can be identi-
fied either establishing a fixed duration or a given number of
tweets.
4.2. Step 2: Semantic structure extraction
The output of the pre-processing step is a dataset composed by
people who sent the messages and keywords they used. The con-
nections between the set of users U and the set of keywords W can
be represented through a 2-mode (“bipartite”) network G ¼ (U, W,
A) such that if ui is a Twitter user and wj is a keyword, there is an
edge aij ¼ ðui ; wj Þ2A if and only if ui sent a tweet containing the
keyword vj . The W-projection of G % GW- is computed to extract the
network of keywords mentioned in the conversation
where GT is the transpose of G. GW represents a network of words
that are connected because they are “shared” by users directly or
through some degree of separation. Those structures are “socially
generated” and they include logic adjacency pairs obtained
through the concatenation of keywords linked together by
different users. The process is cumulative because it supports at-
tempts of expansion or strengthening of shared content: such
content is expanded as users create links to new keywords and
strengthened when associations between keywords are re-
asserted (e.g., through retweets), in which case the weight asso-
ciated to an existing link is increased as the associations becomes
more frequent.
After calculating basic topological metrics for the keywords
network (such as density or centrality), we apply clustering analysis
to extract words clusters that potentially corresponds to topics in
the conversation.
To identify words clusters each we use a method based on based
on k-cores. A k-core H of a graph G is a maximal connected sub-
graph of G in which all vertices have degree at least equal to k. It can
be proved that H is the maximum subgraph with this property
(Alvarez-Hamelin, DallAsta, Barrat, & Vespignani, 2008). The value
of k is determined as follows:
a) the k-core algorithm generates a list of “core values” per each
node and we select as k value the dth largest value in the list
of “core values”, where d is the number of topics to be deter-
mined for each partition. This parameter is proportional to the
total number of keywords in the partition and it can be varied
heuristically to fine tune the granularity of the topic
representation;
b) we then look for clusters into H through the Louvain modularity
community detection method based on greedy optimization
(Black, 2005; Blondel, Guillaime, Lambiotte, & Lefebvre, 2008),
and extract d clusters in each partition;
 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
4.3. Step 3: Reply structure extraction
In Twitter Users may reply to each other using the reply func-
tion, “mentions” of other users in their tweet through special
characters (@ or #) and sometimes the retweet functionality. So it is
possible to trace back the conversational links between any two
users and leverage this information to capture and visualize the
reply structure in tweets stream. The reply structure is a social
network from which a number of social metrics can be extracted. In
particular, we computed the clustering coefficient to check for the
emergence of subgroup of users that are replying to or mentioning
each other in a given time slice. We refer to this metric as relational
clustering coefficient to distinguish it from the clustering coeffi-
cient that is computed over the keywords network for the identi-
fication of topics.
4.4. Step 4: Visualization
The visualization proposed in this paper displays the content
generated by the users, processed through the procedure described
in step 2, as the conversation unfolds along the timeline of the
event of interest (Fig. 1). In particular we are interested in identi-
fying whether the conversation is able to achieve some level of
semantic convergence around a specific topic or related set of
topics of interest for the participants.
Each chunk of tweets that are posted in a given time slice is
represented by a bubble whose diameter is proportional to the
number of messages posted in that time slice. In this way the size of
the bubble highlights the moments in which the conversation is
more crowded. Each bubble contains a white dot at the center, the
larger is this dot is, the higher the number of topical clusters that
are identified in the time slice as described in step 2. In other words,
the white dot is a visual proxy to provide an idea of how much
focused the conversation is in a particular time slice: for instance, a
small dot indicates that participants are very much focused on a
few topics in particular time window.
On the vertical axis we report the semantic similarity among
pieces of the conversation that are contiguous in time. The idea is to
observe if there is any semantic continuity in the stream of tweets
to identify parts of the conversation in which there is potentially
some incremental accumulation of meaning or to visualize points
in which the conversation shifts abruptly to a different subject. In
both cases the analysis of the event as well as of the tweets can be
later used to validate if proximity or semantic jumps are actually
present in the data. The semantic similarity between any two
contiguous bubble is computed as the semantic distance between
the most relevant keywords in each bubble by using Lin similarity
computed over an existing lexical database (Wordnet). First, the
Fig. 1. Example of variable visualizations. First figure shows constant number of tweets with focus of the conversation increasing. Second figure shows the number of tweets
increasing.
787
most relevant keyword in each circle is determined by computing
the similarity between all the words in the cluster through pairwise
comparison; then the keywords are ranked in terms of their
average, intra-cluster similarity. The top keywords are then used to
label the circles and the semantic distance among the representa-
tive labels of two contiguous circles is computed and reported on
the vertical axis in Fig. 1. The Natural Language ToolKit (NLTK) in
Python was used to perform the above calculations.
4.5. Summary
The proposed methodology supports the visualization of
conversational patterns in Twitter backchanneling through the
combination of several analytics that are mined from Twitter
streams. More specifically, we extract topological representations
from the content and the reply structure that is generated by the
users and use a set of topological, traffic and sentiment metrics to
observe how these topologies evolve in order to track what users
are talking about and how they are talking. The proposed visuali-
zations help to summarize this evolution to support media analysts
to develop a better understanding about how users are collectively
making sense of the event. In the specific case study we present in
the next section, the official presentation of new products, the
analysis of the backchanneling behavior can help analysts to un-
derstand on which particular aspects the users focus, if there are
dominant topics or themes, if some topics generate more conver-
sational interaction than others, and how is the sentiment about
specific attributes or features of the new product.
5. Results
This section presents the results obtained by applying the pro-
posed methodology to a corpus of 46,804 tweets in English only,
downloaded in a time window of 1h40’ during the Apple Keynote
Event in March 2015. The tweets were obtained via search through
Twitter API using as keyword “apple”. In order to perform the
search, we created a Python script interfacing Twitter API, down-
loading the tweets over time and storing them in a non-structured
database (MongoDB). We also used our own Python scripts to fully
automate the methodology described in the previous paragraph.
Using those scripts, we processed the tweets we downloaded,
creating the graph in Fig. 2.
Fig. 2 shows the first visualization for the Apple Keynote event
over time. The time axis is segmented into moments corresponding
to the various parts of the presentation as derived from the
verbatim transcript of the event.
A quick visual inspection of the whole graph shows that the
backchanneling conversation taking place on Twitter during the
788 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
Fig. 2. Visualization of the MacBook and AppleWatch release event e The timeline of the event is labeled from the presentation agenda and the transcripts of the keynote speech.
keynote presentation is highly fragmented and scattered; the
variance of the semantic distance is rather high, which implies that
there is, on average, little semantic continuity between adjacent
chunks of tweets. This result seems to reflect the nature of the
event being followed, a keynote in which the speaker is in full
control of the presentation agenda and the speech touches upon a
large variety of topics. The scattered aspect of the conversation,
characterized by abrupt semantic switches between a bunch of
different topics whose content mirrors quite well the keynote
speech content, shows that the conversation is highly reactive and
the users tend to process the event quickly, by sticking to the
speaker's agenda and following the rhythm imposed by the event
timeline.
Fig. 2 also shows that the width of the semantic switches be-
comes quite high each time the keynote speaker shifted to a new
item in the agenda, notably when the speech turned from the Apple
TV to the iPhone, then from the iPhone to the MacBook, and finally
from the MacBook to the AppleWatch.
Despite the scattered character of the overall conversation, it is
possible to notice that the variability of the semantic distance is
different within different macro-topics and that there are some
streams in which the variation between consecutive bubbles is very
small. Here are some examples of “local” smooth conversations:
! launch-major refers to the introduction of a major novelty of the
new the iOS (Car Play, 10:16 am);
! the sequence love-magic represents a first excited reaction to the
introduction of the new MacBook at 10:30 am;
! The sequence why-pushes refers to Apple “obsession” to make
its products thinner and thinner 10:35;
! announcement-want-wants the first reaction of excitement to
the announcement of the Apple watch at 10:53 am;
! sales-pay is related to the possibility to buy items through the
watch with Apple Pay (11:13 am);
! answer-leave refers to Apple Watch functionality that allows its
users to make calls through the watch (11:15 am).
It is interesting to note that the sequence taylor-pro-korean re-
fers to some tweets that we found, upon later analysis, to be
generated by advertisement messages unrelated to the event and
that our platform processed anyway because we decided to not
sanitize the corpus manually before the analysis. The retrieval of
this sequence shows that our algorithm is able to isolate what may
appear to be an unusual sequence to an interpreter that is knowl-
edgeable about the event.
It is interesting to observe that, overall, the average semantic
distance is lower in the MacBook window than for the Apple
Watch; we interpret this finding on the base of the higher famil-
iarity of the users with the MacBook, a new product that, in fact,
exhibited only incremental improvements compared to the previ-
ous model (e.g. better screen resolution, more computational po-
wer, improved battery life, and other minor design changes).
Because of the higher familiarity with the product and the marginal
novelties in the new release, the users were more likely to focus on
a few, related aspects and sustain a less hectic conversation.
Conversely, during the AppleWatch time window, this conti-
nuity is less apparent; we argue that, being the watch a truly new
product, the participants were mainly interested in the “here and
now” and more likely to zigzag from one subject to another to
explore the novel functions and characteristics of the new product,
more or less as they were introduced in the presentation.
Variations in the semantic distance might actually signal two
different types of conversational dynamics: one characterized by
lower variation that is more oriented to exploitation (e.g. more
focus, more in depth discussion, more analysis) and another char-
acterize by higher semantic volatility in which users are exploring,
perhaps in a chaotic and unstructured fashion, the meaning of what
is being said or presented.
Fig. 3 also shows that the bubbles in the AppleWatch portion of
the keynote are bigger than those generated by the updates on the
iPhone and the presentation of the new MacBook. The data shows
that the iWatch attracted more attention, being the real novelty in
the keynote presentation and a product that had been waited for a
long time after several announcements, some deferments, and
well-crafted, anticipated online buzz. The high level of excitement
about the AppleWatch is well represented by the big bubble labeled
“announcement” followed by “want” that appears when the new
device is introduced for the first time.
The users’ focus, which in the proposed visual is given by the
ratio between the area in the white circle and the width of the
bubble, is instead higher in the MacBook case, with the most
focused bubble being love in time slice 174. The difference in focus
 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
Fig. 3. Alternative visualization of the MacBook and AppleWatch release event.
can be interpreted again on the base of the different degree of
novelty of these two products: the MacBook was a new release of a
popular laptop, whose functions and characteristics were sub-
stantially known to the public, so the interest was polarized just on
a few new features and improvements. The AppleWatch was a
totally new product whose applications and target users were
relatively unknown, thus participants were less able to focus on
specific features and uses; on the contrary they seemed more keen
to explore different attributes and potential issues of the new
product.
The highest level of focus during the event was achieved during
the time window from 10:29 a.m.-10:30 a.m US west coast time,
when the MacBook's presentation started. The second highest focus
peak took place during the time window 10:53 a.m.-10:54 a.m.,
when the new smartwatch was introduced to the world. So, these
results suggest that during live events, Twitter users attention is
highly reactive and deeply focused on what they are following.
In Fig. 3 we represent the same event, but this time the size of
the bubbles is proportional to the relational coefficient, a metric
that indicates the presence of more dense conversational clusters in
the tweets stream generated in a given time slice as determined by
the use of reply, mention, and retweet. In other words, the bigger
the bubble, the higher the amount of conversational interaction.
Using this visual it is possible to identify moments and topics in the
event in which the participants engaged more intensively in
conversational exchanges.
Our data in Fig. 3 show that the highest level of conversational
interaction were achieved at the beginning of the event, during the
opening and the phase in which some updates about upcoming
new products were presented (Apple TV, new iPhone). Instead in
the time windows of the MacBook and AppleWatch, the number
and size of conversational clusters drop significantly. By later in-
spection of the tweets' content we found that this behavior can be
explained by the high pace and strong focus of the keynote event:
during the non-core phases such as the opening and the updates,
users were less interested in the event content and had more time
to interact with other participants; during the core phases instead,
the users attention was mainly directed at commenting on the
products’ features. Another factor that may have contributed to
limited conversational interaction could be the information over-
load due to sensible increase in traffic that occurred when the
789
MacBook and the AppleWatch were introduced.
We assessed the trustworthiness of our method by triangulating
the information extracted through the maps with the transcript of
the keynote speech and a set of external sources including blogs
and other media that covered the event and that we assumed as a
sort of baseline (ground truth e Aiello et al., 2013). We found out
that the content extracted through our algorithm was strongly
aligned with the official agenda of the speaker and with what was
echoed by other independent observers of the same event.
6. Discussion and conclusions
6.1. Contribution and implications
In this paper we have presented a methodology to extract and
visualize in an automated way the content generated by Twitter
users in backchanneling conversations. We have presented
empirical findings from a study in which we applied our method to
the analysis of a backchanneling case related to a well-known
event. The method has been developed using as theoretical
perspective based on discourse psychology and conversational
analysis studies to increase our understanding of the process
through which participants use online conversations to make sense
and generate shared representations of the event they are
attending in a virtual fashion. Our findings show that the proposed
method is able to capture and quantify several interesting features
of backchanneling conversational interaction such as: how atten-
tion and focus change across the event, identification and mea-
surement of semantic switches in the conversation, and
identification of moments and topics characterized by the presence
of denser conversational clusters.
By identifying measurable structural properties in the massive
and continuous flow of data that typically characterize SM stream,
our method allows analysts to identify underlying conversational
patterns that could be used to profile conversations and investigate
whether certain configurations are associated to outcomes of in-
terest; for instance, in marketing conversational patterns may be
informative about or even predictive of customers' satisfaction,
purchase intention, or actual consumers' behavior. The analysis of
the “local” conversations can give highly detailed insights about
customers’ concerns or sources of excitement or just point out to
790 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
moments and topics that induced more conversational interaction.
More broadly, conversational patterns may be predictive of in-
dividual or collective behavior, in terms of belief formation, inten-
tion, and individual or collective action (Ajzen, 1991). For instance,
in another study we have found empirical evidence that back-
channeling conversational patterns were predictive of box office
revenues for newly released movies (ref. omitted for anonymity
sake). While there are many studies on the impact of peer to peer
influence on persuasion and behavior in many diverse fields
ranging from marketing (Leskovec et al., 2007) to health behavior
(Centola, 2010), interaction is generally modeled as exchange of
information packages enabled by social links and little empirical
analysis is offered to keep into account the participative, iterative,
and constructive nature of those exchanges.
In particular we think that that more empirical research
assuming conversations as unit of analysis in addition to social
network topology is needed to improve our understanding of how
information and knowledge is created and gained through online
interaction. For instance some conversational patterns may be
associated to more informative interaction that could possibly lead
to individual and collective learning. The study of learning in virtual
social networks based on conversational analytics could provide
deeper insight on how online conversation and other collaborative
tool can be used and also improved to facilitate learning and
improve quality of online debate.
From the practical point of view, the proposed methodology
could support marketing and opinion analysts in better managing
the reach-richness tradeoff that is typical of marketing or social
research. Traditionally, marketing analysts use qualitative research
methods, such as focus groups, applied to small samples of pro-
spective or actual customers to gain an in-depth understanding of
the factors influencing customers' preferences and of the causal
relationships between product attributes, product performances,
and customers’ needs. However traditional qualitative methods
have two major drawbacks: they require subjective interpretation
of the findings and cannot be applied to large sample because
qualitative analysis is very costly and time consuming. On the other
hand, large-scale, quantitative traditional market analysis, based on
standard surveys tends to be shallow, especially when it comes to
the identification of emerging/unexpected trends.
More broadly the proposed method could be used in other ap-
plications in which analysts or decision makers want to learn from
an event for which backchannel data are accessible almost in real
time, such as in emergency management (Hughes & Palen, 2009),
political mobilization (Theocharis, Lowe, van Deth, & García-
Albacete, 2015) or live events (Ross, Terras, Warwick, & Welsh,
2011).
6.2. Limitations and future developments
We decided to use Twitter because of its popularity and con-
venience. The proposed method could be applied to other more
popular SM through the appropriate changes to the data ingestion
step; with more conversational venues such as online forums the
linguistic and semantic complexity of conversations in which users
are not constrained to use very short messages would require more
sophisticated semantic preprocessing than the relatively simple,
lexical approach that we present in this paper. Moreover, user-
generated content in SM such as Facebook does not have the
same level of accessibility because users tend to not share their
content publicly and Facebook endorses a less liberal data access
policy.
A limitation of the present study is that the observation interval
is limited to an event of less than 2 h. While the short duration
helps to create focus, it may prevent the observation of more long-
term, self-organized patterns that might be generated in SM that
are supportive of more sophisticated communication, such as
Facebook.
One inconvenience of using SM data is that the sample of users
may not be statistically representative. This problem could be
overcome at least in part by extracting demographics from users’
online profile or through the set-up of ad hoc, closed discussions in
which participants are recruited in an unbiased way. However, even
demographic controls could not overcome the fact that the audi-
ence of the event is self-selected: the Apple event is a case in point
because it is reasonable to assume that Apple fans were over-
represented among the participants who followed the keynote.
This may explain the high level of excitement for the AppleWatch
that was probably behind the record early sales of the product, but
that proved to not be enduring as the subsequent sales level
dropped significantly and were later judged as disappointing from
market analysts. A possible way to escape the self-selection bias is
to triangulate the findings obtained through SM mining with those
obtained through more neutral and traditional data collection
methods.
Another critical issue in this respect is the increasing use of e-
bots publishing tweets that are generated by machines. The
detection of these manipulation attempts is not easy to perform,
was outside the scope of this work and we are aware that several
researchers are developing algorithms to automatically detect
spamming and other artificial ways to bias the conversation (Zhao
et al., 2014). However it was comforting for us to find out that
despite the lack of countermeasure to spam attempts, our algo-
rithm was able to isolate one local conversation that was originated
by unrelated advertisement. Thus, a future development of our
work may concern its application to the detection of possible spam
based on semantic analysis, very much in the same way as human
analysts could spot incoherent segments of conversation.
Our initial validation allowed us to pass a face validity test and
achieve some degree of external validity through the comparison of
the maps generated by our tool with content on the same event
available through independent sources. A more structured and
rigorous topic validation method is of course required, but given
the complexity and methodological nuances involved in this type of
assessment and the space constraint, it was not possible to include
a detailed account of the validation methodology in this paper.
These materials are however available to interested readers on
request.
We collected tweets with no language restriction but we only
considered English posts in the analysis. This is a limitation for
events that have an international audience. On the other hand,
because we were interested in spotting conversational patterns and
because people need to speak the same language to have a con-
versation, we think that the non inclusion of tweets in other lan-
guage was not relevant for our purposes. Besides, our method can
be applied to other languages by using the appropriate dictionary.
Since our method is fully scalable, further replications of the
proposed analysis over a large sample of comparable events could
be desirable to identify statistical correlation between some of the
metrics we propose and provide stronger empirical validation of
the method; however, the data collection for a large sample of
similar events is time consuming and very expensive, especially if
the analysis requires the acquisition of past data from professional
data providers, whose costs may prove to be unaffordable to most
academic budgets. For this reason we preferred to focus on single
case study to show the potentiality and usability of our method-
ology and provide some validation based on empirical findings
from real event. In a parallel line of work, we are applying our
analysis to a larger sample of conversations to find out if the
conversational patterns elicited through our method can help to
 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
predict consumers behavior in the purchasing of newly launched
products.
Acknowledgements
The research leading to these results has received funding from
the Strategic Research Counsel at the Academy of Finland under
grant agreement n. 293446 e Platform Value Now: Value capturing
in the fast emerging platform ecosystems.
References
Aiello, L. M., Petkos, G., Martin, C., Corney, D., Papadopoulos, S., Skraba, R., et al.
(2013). Sensing trending topics in Twitter. IEEE Transactions on Multimedia,
15(6), 1268e1282.
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human
Decision Processes, 50(2), 179e211.
Alvarez-Hamelin, J. I., DallAsta, L., Barrat, A., & Vespignani, A. (2008). K-Core
decomposition of Internet graphs. Networks and Heterogeneous Media, 3(2),
1e23.
Baraba!si, A. L., Albert, R., & Jeong, H. (2000). Scale-free characteristics of random
networks: The topology of the world-wide web. Physica A: Statistical Mechanics
and its Applications, 281(1), 69e77.
Bartlett, J., Froio, C., Littler, M., & McDonnell, D. (2013). New political actors in Europe:
Beppe Grillo and the M5S.
Bauckhage, C. (2011). Insights into Internet memes. In International AAAI conference
on web and social media (pp. 42e49).
Black, P. E. (2005). Dictionary of algorithms and data structures at the National
Institute of standards and Technology (NIST).
Blondel, V., Guillaime, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of
communities in large networks. Journal of Statistical Mechanics: Theory and
Experiment, 2008(10), 1e12.
Bolton, R. N., Aksoy, L., van Riel, A., & Kandampully, J. (2013). Understanding gen-
eration Y and their use of social media: A review and research agenda. Journal of
Service Management, 24(3), 245e267.
Brachman, J. (2014). Transcending Organization: Individuals and 'the Islamic state.
College Park, Maryland: START Analytical Brief.
Brown, E. D. (2012). Will twitter make you a better investor? a look at sentiment,
user reputation and their effect on the stock market. In Conference proceedings
for the southern association for information systems SAIS conference (pp. 36e42).
Burt, R. S. (2004). Structural holes and good ideas. American Journal of Sociology,
110(2), 1e51.
Centola, D. (2010). The spread of behavior in an online social network experiment.
Science, 329(5996), 1194e1197.
Chamlertwat, W., Bhattarakosol, P., & Rungkasiri, T. (2012). Discovering consumer
insight from twitter via sentiment analysis. Journal of Universal Computer Sci-
ence, 18(8), 973e992.
Clark, H. H. (1996). Using language (vol. 4). Cambridge: Cambridge University Press.
Clark, H. H., & Brennan, S. E. (1991). Grounding in communication. Perspectives on
Socially Shared Cognition, 13(1991), 127e149.
Clark, H. H., & Schaefer, E. F. (1989). Contributing to discourse. Cognitive Science,
13(2), 259e294.
Croft, W. B., Metzler, D., & Strohman, T. (2010). Search engines: Information retrieval
in practice (vol. 283). Addison-Wesley, Ed.
Cui, W., Liu, S., Tan, L., Shi, C., Song, Y., & Gao, Z. (2011). TextFlow: Towards better
understanding of evolving topics in text. IEEE Transactions on Visualization and
Computer Graphics, 17(12), 2412e2421.
Culnan, M., McHugh, P., & Zubillaga, J. (2010). How large US companies can use
Twitter and other social media to gain business value. MIS Quarterly Executive,
243e259.
Dork, M., Gruen, D., Williamson, C., & Carpendale, S. (2010). A visual backchannel for
large-scale events. Visualization and Computer Graphics. IEEE Transactions on,
16(6), 1129e1138.
Duggan, M. (2015, August 19). The demographics of social media users. Retrieved
from http://www.pewinternet.org/2015/08/19/the-demographics-of-social-
media-users/.
Evangelopoulos, N., Magro, M. J., & Sidorova, A. (2012). The dual micro/macro
informing role of social network Sites: Can twitter macro messages help predict
stock prices? Informing Science: the International Journal of an Emerging Trans-
discipline, 15, 247e267.
Gartner. (2013). Hype cycle for social software, 2013. Retrieved from http://www.
gartner.com/newsroom/id/2575515.
Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology,
78(6), 1360e1380.
Greenhow, C., & Greenhow, C. (2011). Online social networks and learning. On the
Horizon (vol. 19(1), pp. 4e12).
Harrington, S., Highfield, T., & Bruns, A. (2012). More than a backchannel: Twitter
and television. Audience Interactivity and Participation, 13e17.
He, W., Zha, S., & Li, L. (2013). Social media competitive analysis and text mining: A
case study in the pizza industry. International Journal of Information Manage-
ment, 33(3), 464e472.
791
Highfield, T., & Bruns, A. (2013). More than a backchannel: Twitter and television.
Journal of Audience Reception Studies, 1e5.
Hoffman, D., & Fodor, M. (2010). Can you measure the ROI of your social media
marketing? (pp. 1e11).
Hughes, A. L., & Palen, L. (2009). Twitter adoption and use in mass convergence and
emergency events. International Journal of Emergency Management, 6(3e4),
248e260.
Hu, X., & Liu, H. (2012). Text analytics in social media. In Mining text data (pp.
385e414). Boston, MA: Springer US.
Introne, J. E., & Drescher, M. (2013). Analyzing the flow of knowledge in computer
mediated teams (p. 341). Presented at the CSCW '13: Proceedings of the 2013
conference on Computer supported cooperative work, New York, New York,
USA.
Jaegher, H. D., & Paolo, E. D. (2007). Participatory sense-making. Phenomenology and
the Cognitive Sciences, 6(4), 485e507. http://doi.org/10.1007/s11097-007-9076-
9.
Klausen, J. (2015). Tweeting the Jihad: Social media networks of Western foreign
fighters in Syria and Iraq. Studies in Conflict & Terrorism, 38(1), 1e22.
Leskovec, J., Adamic, L. A., & Huberman, B. A. (2007). The dynamics of viral mar-
keting. ACM Transactions on the Web (TWEB), 1(1), 5.
Manning, C., Raghavan, P., & Shutze, H. (2009). An introduction to information
retrieval (pp. 1e569). Cambridge: Cambridge University Press.
McNely, B. (2009). Backchannel persistence and collaborative meaning-making. In
Proceedings of the 27th ACM international conference on Design of communication
(pp. 297e304). ACM.
Melville, P., Sindhwani, V., & Lawrence, R. (2009). Social media analytics: Chan-
neling the power of the blogosphere for marketing insight. Proc. of the WIN,
1(1), 1e5.
Mills, A., Chen, R., Lee, J., & Raghav Rao, H. (2009). Web 2.0 emergency applications:
How useful can Twitter be for emergency response? Journal of Information
Privacy and Security, 5(3), 3e26.
Moscovici. (2001). Social Representations: Explorations in social psychology. New
York: NYU Press.
Naveed, N., Gottron, T., Kunegis, J., & Alhadi, A. C. (2011). Searching microblogs:
coping with sparsity and document quality (pp. 183e188). Presented at the
CIKM '11: Proceedings of the 20th ACM international conference on Informa-
tion and knowledge management, New York, New York, USA.
Nielsen. (2013, August 6). New Nielsen research indicates two-way causal influence
between Twitter activity and TV viewership. Retrieved February 14, 2014, from
http://wp-s3.socialguide.com.s3.amazonaws.com/wp-content/uploads/2013/
08/080613.NielsenTwitterCausation.FINAL_.pdf.
Parker, E. (2014). Social media and the Hong Kong protest. The New Yorker magazine,
October 1st.
Perrin, A. (2015). Social media usage: 2005-2015. Retrieved from http://www.
pewinternet.org/2015/10/08/social-networking-usage-2005-2015/.
Potter, J., & Edwards, D. (1999). Social representations and discursive psychology:
From cognition to action. Culture & Psychology, 5(4), 447e458.
Pousman, Z., Stasko, J. T., & Mateas, M. (2007). Casual information Visualization:
Depictions of data in everyday life. IEEE Transactions on Visualization and
Computer Graphics, 13(6), 1145e1152.
Qian, W., Zhou, A., Zhang, Z., & Zhao, B. (2013). Identification of collective view-
points on microblogs. Datak, 87(C), 374e393.
Rogers, E. M. (1995). Diffusion of innovations. Free Pr.
Ross, C., Terras, M., Warwick, C., & Welsh, A. (2011). Enabled backchannel: Con-
ference Twitter use by digital humanists. Journal of Documentation, 67(2),
214e237.
Sack, W. (2000). Conversation Map: An Interface for very large-scale conversations.
Journal of Management Information Systems, 17(3).
Saif, H., He, Y., & Alani, H. (2012). Semantic sentiment analysis of twitter (Vol. Part I,
Volume Part I). In Presented at the ISWC'12: Proceedings of the 11th international
conference on the semantic web. Springer-Verlag.
Shamma, D. A., Kennedy, L., & Churchill, E. F. (2009). Tweet the debates: Under-
standing community annotation of uncollected sources. In Proceedings of the
first SIGMM workshop on Social media (pp. 3e10). Retrieved from http://dl.acm.
org/citation.cfm?id¼1631148.
Siemens, G. (2005). Connectivism: A learning theory for the digital Age. Interna-
tional Journal of Instructional Technology and Distance Learning. Retrieved from
http://er.dut.ac.za/handle/123456789/69.
Sloane, N., & Wyner, A. (1948). A mathematical theory of communication. Claude E.
Shannon: Collected Papers (pp. 5e83).
Smith, M. A., Rainie, L., Shneiderman, B., & Himelboim, I. (2014). Mapping twitter
topic networks: From polarized crowds to community clusters. Retrieved from
Pew Research Internet Project http://www.pewinternet.org/2014/02/20/part-2-
conversational-archetypes-six-conversation-and-group-network-structures-in-
twitter/.
Sysomos. (2010). Replies and retweets on twitter. Retrieved from https://sysomos.
com/inside-twitter/twitter-retweet-stats.
Theocharis, Y., Lowe, W., van Deth, J. W., & García-Albacete, G. (2015). Using twitter
to mobilize protest action: Online mobilization patterns and action repertoires
in the occupy wall street, indignados, and Aganaktismenoi movements. Infor-
mation, Communication & Society, 18(2), 202e220.
Tsur, O., & Rappoport, A. (2012). What's in a hashtag?: content based prediction of
the spread of ideas in microblogging communities (p. 643). Presented at the
WSDM '12: Proceedings of the fifth ACM international conference on Web
search and data mining, New York, New York, USA.
792 C. Lipizzi et al. / Computers in Human Behavior 64 (2016) 782e792
Uzzi, B. (1997). Social structure and competition in interim networks. Administrative
Science Quarterly, 1e34.
Vespignani, A. (2005). Complex networks: Behind enemy lines. Nature Physics, 1(3),
135e136.
Vie
!gas, F. B., Golder, S., & Donath, J. (2006). Visualizing email content: Portraying
relationships from conversational histories. In Proceedings of the SIGCHI con-
ference on Human Factors in computing systems (pp. 979e988).
Wakamiya, S., Lee, R., & Sumiya, K. (2011). Towards better TV viewing rates:
exploiting crowd's media life logs over Twitter for TV rating (p. 1). Presented at
the ICUIMC '11: Proceedings of the 5th International Conference on Ubiquitous
Information Management and Communication, New York, New York, USA.
Williams, D. L., Crittenden, V. L., Keo, T., & McCarty, P. (2012). The use of social
media: An exploratory study of usage among digital natives. Journal of Public
Affairs, 12(2), 127e136.
Yu, A. Y., Tian, S. W., Vogel, D., & Kwok, R. C.-W. (2010). Can learning be virtually
boosted? an investigation of online social networking impacts. Computers &
Education, 55(4), 1494e1503.