Journal of Retailing and Consumer Services xxx (xxxx) xxxx

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services

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

A typology of viral ad sharers using sentiment analysis

Kalpak K. Kulkarnia, , Arti D. Kalroa, Dinesh Sharmaa, Piyush Sharmab
⁎

a
SJM School of Management, Indian Institute of Technology Bombay, Mumbai, India
b
School of Marketing, Curtin Business School, Curtin University, Bentley, WA, Australia

ARTICLE INFO ABSTRACT

Keywords: Viral advertising is the most popular manifestation of viral marketing phenomena. The purpose of this study is to
Viral advertising demonstrate sentiment analysis as a promising tool to quantify consumer responses towards branded viral video
Cognitive responses advertisements and thereupon, propose a sentiment-based typology of viral ad sharers. Results of this experi-
Thought-listing method mental study (1) suggest that sentiment-based measures of consumer responses offer better prediction of con-
Sentiment analysis
sumers’ ad sharing intentions compared to the traditional and widely used thought-listing method; and (2) help
Typology
identify four distinct segments of viral ad sharers (based on the relative strength of ad- and brand-related
sentiments), namely: “Active”, “Brand-fanatic”, “Content-hungry”, and “Dormant”, labeled as ABCD typology of
viral ad sharers. This study highlights that for creating successful viral campaigns, marketers should consider the
distinctive characteristics of these four segments of viral ad sharers (based on their processing of ad content and
brand information) to identify the right seeds to initiate a viral campaign.

1. Introduction advertising messages and seed them in video platforms encouraging

consumers to share these messages with their online social networks
In this age of the Internet, social media has become one of the most (Berger and Iyengar, 2013). World-renowned brands, such as Burger
powerful branding tools for organizations to effectively communicate King, Evian, Old Spice and Unilever's Dove have successfully employed
with their targeted customers and enhance the impact of promotional viral advertising (Kaplan and Haenlein, 2011; Beverland et al., 2015), a
activities on the customers’ perceptions and awareness levels (Alalwan new and promising marketing communications’ tool to engage millions
et al., 2017; Nisar et al., 2018; Alalwan, 2018). From the consumers’ of consumers instantly (Petrescu et al., 2016).
perspective, social media and social networks have become a part of Consumers’ processing of viral ads follows the popular ‘Emotion-
their daily lives (Shareef et al., 2018, 2017; Shiau et al., 2017, 2018) Cognition Model’ (Zajonc and Markus, 1982), wherein viral ads are
and these have changed the way in which individuals acquire in- found to elicit emotions and trigger the cognitive appraisal that moti-
formation and communicate with each other (Dwivedi et al., 2015; vates consumers to pass-on the ad to their friends and connections
Kamboj et al., 2018; Kapoor et al., 2018; Aswani et al., 2018). In the last (Berger and Milkman, 2012; Huang et al., 2013). At present, marketers
decade, significant interest in social media marketing has been seen in don’t have direct access and/or appropriate tools to measure emotions
terms of advertising from both researchers and practitioners (Alalwan of every consumer viewing their ad; however, the appraisal of such
et al., 2017; Shareef et al., 2017, 2018; Kamboj et al., 2018; Alalwan, emotions does get captured in the cognitive responses that are easily
2018). This is also evident considering the global video advertising available on social media platforms in the form of user generated
market size that has grown to about US$28 billion (as of 2018) and is content (such as textual responses/comments on the ads). In the context
further expected to grow at an annual rate of 14.6% (Statista, 2017). of viral ads, Huang et al. (2013) showed that such cognitive responses,
Kantar Millward Brown (2018) report titled ‘Digital and Media Pre- generated in response to a viral ad, trigger formation of subsequent
dictions’ highlights that online video advertisements are the most sig- attitude and sharing intention (SI). Although there are several studies
nificant form of content marketing because of their expressiveness and that have used the traditional ‘thought-listing method’ (TLM) to analyze
their ability to elicit active involvement of the viewers. Particularly, cognitive responses (Wright, 1973; Cacioppo and Petty, 1981) across
online viral branding campaigns (viral advertising) are an emerging different topics like consumer evaluation of online banner advertising
trend, wherein advertisers create attractive and entertaining (Yun Yoo and Kim, 2005), Facebook advertising (Shareef et al., 2018),

Corresponding author.
⁎

E-mail addresses: [email protected] (K.K. Kulkarni), [email protected] (A.D. Kalro), [email protected] (D. Sharma),
[email protected] (P. Sharma).

https://doi.org/10.1016/j.jretconser.2019.01.008
Received 14 September 2018; Received in revised form 10 January 2019; Accepted 10 January 2019
0969-6989/ © 2019 Elsevier Ltd. All rights reserved.

Please cite this article as: Kalpak K. Kulkarni, et al., Journal of Retailing and Consumer Services,
https://doi.org/10.1016/j.jretconser.2019.01.008
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

as well as sharing of viral advertisements (Huang et al., 2013); to the on products (Schulze et al., 2014) or brand characteristics (Lovett et al.,
best of our knowledge, there has been no study that has analysed these 2013) identify key features of product categories or brands that drive
cognitive responses to segment consumers for seeding viral ads. word-of-mouth (WOM) and motivate people to talk about them. Per-
Moreover, this thought-listing method has serious limitations con- taining to content, existing studies show that consumers share messages
cerning empirical indicators of measurements (Lutz and Swasy, 1977; that offer high entertainment and enjoyment levels (Phelps et al.,
Huang and Hutchinson, 2008). This method captures consumers’ eva- 2004), high utilitarian or hedonic value (Chiu et al., 2007), or high
luations of objects in the valence-congruent direction (Olson et al., levels of emotional experience (Dobele et al., 2007; Berger and
1982), but fails to appraise the effect of the intensity of responses on Milkman, 2012). Studies on recipients/audience of these messages seek
consumers’ attitudes and behaviours. Nevertheless, areas of psycho- to identify people (seeds) with a higher propensity to share, based on
linguistics and Natural Language Processing (NLP) have introduced their personality traits (Chiu et al., 2007), motivation (Ho and
tools like sentiment analysis that can effectively determine the valence Dempsey, 2010), or positions in a social network (Hinz et al., 2011).
as well the strength of consumers’ opinions. Hence, we advocate the use Very few studies have looked at seeding strategies, that is, the
of sentiment analysis for a holistic measurement of these cognitive re- careful selection of the initial target consumers and placement of a viral
sponses and use these sentiment-based measures of cognitive responses message (Kiss and Bichler, 2008; Zhang et al., 2013). For a successful
to further segment consumers/sharers to help marketers identify the seeding strategy, Hinz et al. (2011) identified four decisive factors:
“actionable segment” of consumers or ‘seeds’, who show a higher content (e.g., Berger and Schwartz, 2011), network structure (e.g., Kiss
likelihood of sharing viral ads vis-à-vis other consumers. This may help and Bichler, 2008), behavioural incentives (e.g., Libai et al., 2013) and
marketers effectively design and distribute advertising content using the seeding strategy itself (e.g. Liu-Thompkins, 2012). Research in-
the viral phenomenon. With this background, an experimental study is vestigating optimal seeding strategies has largely focused on the issue
designed to address the following two research objectives: relating to the profiles of the ‘seeded’ individuals, often labeled as
‘hard-to-find’ influencers, who have a disproportionate effect on the
1. To examine whether sentiment analysis is a better tool compared to ‘others’ (Trusov et al., 2010). These influencers can be identified based
the traditional thougth-listing method (TLM) in explaining the on either their social network features (Kiss and Bichler, 2008; Zhang
Social Networking Site (SNS) users’ viral ad sharing intentions? et al., 2013) or certain psychological variables like personality traits
2. To develop a typology of viral ad sharers using sentiment-based (Chiu et al., 2007). However, individuals differ in their information
measures of cognitive responses. processing abilities (Petty and Cacioppo, 1986), which can be measured
by analysing their cognitive responses. In their study, Huang et al.
This study contributes to the existing literature by demonstrating (2013) found that ad- and brand-related cognitions (Cad, Cbr) drive
the usefulness of sentiment analysis to overcome the measurement sharing intentions (SI) for viral ads. However, no attempt has been
problems associated with the traditional thought-listing method used made to analyze these consumer cognitions in the form of unstructured
for analysing consumers’ cognitive responses generated in response to (latent) texts as a means to identify and segment influencers (initial
viral advertisements. There are few important studies that have ex- sharers) of viral ads. Researchers studying cognitive responses have
plored content virality across different social media platforms [like largely benefited from the ‘Cognitive Response Theory’ for con-
Facebook (Aswani et al., 2017a), Twitter (Aswani et al., 2017b) and ceptualizing and measuring these responses. This theory has been dis-
online news articles (Aswani et al., 2017c)] to identify what type of cussed in the following section.
content goes viral or becomes popular. By taking the context of viral
video advertisements, this study contributes to the above literature by 2.2. Cognitive Response Theory (CRT)
identifying types of ‘consumers’ that can actually trigger/stimulate
content virality. Further, there exists number of typologies to segment Cognitive responses are simply the thoughts that consumers con-
Internet or social media users (e.g. types of Facebook Fans as defined by struct in response to persuasive advertising (Petty et al., 2014). Ac-
Wallace et al., 2014) that help marketers strategise their marketing cording to the Cognitive Response Theory, spontaneous thoughts
efforts and target the appropriate set of consumers. However, there (cognitive responses) elicited by an exposure to a message act as direct
remains a lack of such segmentation to classify viral ad sharers. Using mediators of attitude formation or change (Greenwald, 1968; Petty
the psychographic segmentation approach, this study proposes a ty- et al., 2014). According to Wright (1973), cognitive responses can be
pology of viral ad sharers that will help marketers identify appropriate further classified into: counterarguments (CA), support arguments (SA),
‘seeds’ to begin their branding campaigns and make them viral. source derogation (SD) and source bolstering (SB). Among these, CA
This paper is structured as follows: First, we review the literature on and SD are believed to result in less favourable attitudes, whereas SA
seeding strategies in viral advertising, Cognitive Response Theory and SB result in positive attitudes.
(CRT) and sentiment analysis. Next, in the Methodology section, we
demonstrate the effectiveness of sentiment analysis over the traditional 2.2.1. Measurement of cognitive responses
thought-listing method in measuring consumer cognitions. The traditional method of measuring cognitive responses to an ad-
Subsequently, using ad- and brand-related sentiments, we con- vertisement is the thought-listing method (e.g., Cacioppo and Petty,
ceptualize and illustrate a typology of viral ad sharers. Finally, we 1981; Wright, 1973; Huang and Hutchinson, 2008), in which the re-
conclude with theoretical and managerial implications, limitations of spondents are asked to write down all their thoughts either during/
this study, and future areas for research. immediately after exposure to a message. Then, the cognitive responses
are categorized according to the various criteria by the subject/ex-
2. Literature review perimenter to see whether or not the responses meet the prescribed
definition for a class of cognitive responses [for example, CA, SA]
2.1. Viral advertising and seeding strategies (Wright, 1973, 1974). The third step is to compute the cognitive re-
sponse scores that represent either a simple addition of the number of
Viral potential of advertising messages is a key benefit of using CA, SD, or SA, or more elaborately, use a model in which each cognitive
Internet marketing (Eckler and Bolls, 2011; Porter and Golan, 2006). response type is weighed by subjective indications of importance to
Academicians (Nelson-Field et al., 2013) and practitioners (Valos et al., predict attitude or intention. Typically, these cognitive responses are
2010) suggest a need to understand why only a few viral campaigns coded for valence and summed or averaged to form a measure of the net
(less than 5%) succeed. Extant viral advertising research mainly focuses affective response. This measure is a reliable predictor of attitudes (e.g.,
on products, content or the recipient of these ads. For example, studies Petty and Cacioppo, 1979; Wright, 1973). To the best of our knowledge,

2
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

the thought-listing method has always been used to compute a valence-

weighted aggregate index of cognitive responses, and no measures of
specific thoughts have been used to predict attitudes. Moreover, TLM-
based indicators of cognitions have potential measurement problems.
This is because they are proposed on an intuitive, ad hoc basis to ca-
tegorize cognitive responses at a broad, fairly abstract level, typically in
terms of the evaluative direction of the thoughts vis-à-vis that of the
message (Lutz and Swasy, 1977), capturing the dimension of ‘direction’,
but failing to measure the ‘strength or intensity’ of cognitive thoughts
(Greenwald, 1968). Furthermore, TLM indicators are subjective in
nature, making them unsuitable to study the subtle nuances in the
message on attitudes and intentions. Thus, for an objective assessment
of cognitive responses (Lutz and Swasy, 1977), an improvised method is
needed to quantify these cognitive responses in terms of both, their Fig. 1. Research methodology – thought-listing method vs. sentiment analysis.
valence/direction and strength/intensity of the valence. This is where
sentiment analysis, a method from psycho-linguistics is more appro-
priate to overcome the issues associated with the traditional thought-
listing method. In the next section, we discuss sentiment analysis in author screened the “TOP 100” viral videos from one of the top video
detail. ranking website ‘Viral Video Chart™’ (Powered by Unruly®) to identify
ads that (1) have high ratings and a large number of views, (2) contain
2.3. Sentiment analysis product and brand information and (3) contain a complete story and
were not more than five minutes in duration. After initial screening,
Sentiment analysis, also known as opinion mining, is a field of study seventy viral ads were selected, which were then screened by two
that analyses people's opinions, sentiments, evaluations, appraisals, trained coders (students), each picking the top 20 of their favourite
attitudes, and emotions towards entities such as products, services, videos and they also categorised these ads as either ‘emotional’ or ‘ra-
organizations, individuals, issues, events, topics, and their attributes tional/informational’. On cross-comparison of these “Top 20″ viral ads
(Pang and Lee, 2008; Aswani et al., 2017a). Consumers are able to of both the coders, ten common videos were selected. Then these
evangelize their own experience with many more people via the social shortlisted ten videos were shown to another pool of fifty-six re-
media (Alalwan et al., 2017). Hence businesses are increasingly cap- spondents. After viewing the ad, each participant was asked to report
turing more information about their customers’ sentiments that has led their sharing intention for these ads on a 7-point Likert scale (Huang
to wide applications of sentiment analysis in various business domains et al., 2013), based on which the top two viral ads, one emotional
including marketing (Liu, 2010; Dwivedi et al., 2015). There are dif- (Google Search: Reunion; SI = 6.41) and one rational (Volvo Trucks -
ferent types of sentiment analysis techniques, such as feature-based The Epic Split, SI = 5.50) were selected for the experiment (See
sentiment analysis and document-based sentiment analysis. Feature- Appendix A for the total counts on the actual views and shares of the
based techniques are mostly used to extract a special feature from a shortlisted ads on different social media platforms). A pre-test con-
piece of textual document; whereas, document-based techniques are firmed that the two shortlisted viral ads were perceived differently
more appropriate to extract the overall polarities of a piece of text (MVolvo = 2.85 vs. MGoogle = 4.40; t (34) = 6.46, p < .01) in terms of
(Eslami and Ghasemaghaei, 2018; Liu, 2010). Sentiment analysis, a ad appeal (Liu and Stout, 1987) by the different set of undergraduate
social media analytics tool (Misirlis and Vlachopoulou, 2018; Rathore participants (n = 35; 57.1% males; average age = 21 years) who were
et al., 2017), can be used to extract the polarity of the consumer opi- not the part of the final experiment; wherein the Volvo ad was rated
nions in terms of positive, negative, or neutral and also assign a score to high on the rational and the Google ad was rated high on the emotional
reflect the strength of the sentiment (Kim et al., 2016). Hence, such appeal.
tools seem to be helpful in measuring both, the valence as well as the
‘strength’ of cognitive responses. Thus, we propose that sentiment 3.2. Participants and procedure
analysis is an effective alternative to the TLM in measuring and quan-
tifying the affective strength of consumers’ cognitive responses, that Predominant social network site (SNS) users are young adults under
plays an important role in consumers’ attitude formation and sharing the age of 25 years (Correa et al., 2010) and they are the most active
intentions for the viral ads. video sharers (Santos et al., 2007). Particularly, college students often
are the key target segment for viral advertising campaigns (Chen and
3. Methodology Lee, 2014; Chu, 2011; Lee and Hong, 2016). Hence, the experiment was
conducted with undergraduate and postgraduate students (n = 344;
We designed an experiment to empirically compare and test whe- 72.1% males; average age = 20 years, range = 18–25 years) enrolled
ther sentiment analysis is a more effective method to understand cog- at a large public university in India. Among them, 194 participants
nitive responses better than the traditional thought-listing method. At watched the ‘Google’ ad (https://youtu.be/gHGDN9-oFJE) and 150
first, cognitive responses were collected and analysed for their influence watched the ‘Volvo’ ad (https://youtu.be/M7FIvfx5J10), before filling
on attitudes towards the ad and the brand and sharing intentions, using the questionnaire.
both the thought-listing (Model I) and the sentiment analysis (Model II)
methods. This further helped in selecting the method that offered a 3.3. Measures
richer understanding of consumers’ cognitions (See Fig. 1) and effec-
tively segmenting viral ad sharers. Brand attitude (Abr) was measured using a three-item, 7-point se-
mantic differential scale: good/bad, attractive/unattractive, and high
3.1. Stimuli quality/low quality (Huang et al., 2013; Olson et al., 1982). Measure-
ment of attitude toward the viral advertisement (Aad) included a four-
Viral ads from three years prior were chosen to avoid recent ex- item, 7-point semantic differential scale: like/dislike, favourable/un-
posure effects. Following the procedures of Huang et al. (2013), the first favourable, interesting/boring, and extraordinary/ordinary (Huang

3
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

et al., 2013). Sharing Intention (SI) was measured using a three-item, 7- significant predictors of SI (p < .01) under both the TLM- and the
point scale that captures the extent to which a respondent feels s/he sentiment-based approaches. These results illustrate that the sentiment-
will share the advertisement (1 being the least and 7 being the highest): based measures are more effective (Model II; Adj. R2 = 0.189) in
pass-along probability, probability of telling others, and probability of predicting sharing intentions, when compared to the traditional TLM-
talking about the video (Huang et al., 2013). All variables reported based indicators (Model I; Adj. R2 = 0.140).
acceptable reliability values (Cronbach's α > 0.70) for both the ads. To provide a rigorous test of the differential predictiveness of the
Consumers’ ad- and brand-related cognitions were analysed using two methods, we performed a set of regression analyses as suggested by
the TLM as well as with a sentiment analysis tool, Semantria®. Sirgy et al. (1997). The first set entered the traditional measures (Cad,
Cbr) into the regression equation (R2 = 0.027), followed by the new
3.3.1. Thought-listing Method (Model I) measures (Sad and Sbr) (R2 = .101). If the hypothesis (that the new
We adopted the open-ended Thought-listing Method (Cacioppo and measure is more predictive than the traditional measure) is true, then
Petty, 1981) to measure ad- and brand-cognition responses (Huang we should expect the R2 change due to the addition of the new mea-
et al., 2013). The thought-listing instructions were as follows “please sures to be significant. In this case, the R2 change was significant (R2
write down all the thoughts, ideas, and images that occurred to you change = 0.074; p < .001). Conversely, if we enter the new measures
about the ad and brand while viewing the video”. An independent coder (Sad and Sbr) first (R2 = 0.090), followed by the traditional measure
collected all the responses and sorted them into brand- or video-related i.e. Cad and Cbr (R2 = 0.101), we should expect R2 change to be non-
thoughts, global evaluation thoughts, and irrelevant thoughts. Global significant. This was evident here (R2 change = 0.011, p > .10). As
evaluation and irrelevant thoughts were not considered for further the range of scores for the sentiment analysis was greater (-1.5 to +1.5)
analysis. Then, two trained independent coders coded the responses than that for the thought-listing method (-1 to +1), we repeated the
into positive (+1), neutral (0), or negative opinions (-1). Out of the analysis by using the same range of scores for sentiments analysis as
total 688 opinions coded, the inter-coder reliability coefficient, Krip- well as TLM and found that the results were the same. Hence, this
pendorff's alpha was 88% in case of ad-related cognitions (Cad) and provides additional support for the effectiveness of sentiment analysis
92% for brand-related cognitions (Cbr). A third coder was consulted to method over the traditional TLM, when measuring cognitive responses.
resolve the discrepancies. Therefore, the sentiment-based cognitive measures were selected for
conducting the cluster analysis for testing whether any segments of
3.3.2. Sentiment analysis (Model II) viral ad sharers can be identified based on their sentiments expressed
There are different types of sentiment analysis techniques, such as on the ads.
feature-based sentiment analysis and document-based sentiment ana-
lysis (Eslami and Ghasemaghaei, 2018; Liu, 2010). In this study, as we 4.2. Cluster analysis
were interested in the consumer evaluation of the video ad content and
the brand shown in the viral ads, we used the document-based senti- The ANOVA results revealed that both ad-related sentiments (Sad)
ment analysis method; wherein each specific thread of the ad- and {MLowSad = 5.15, MHighSad = 5.97; F (1, 343) = 32.02, p < .01} and
brand-related thoughts were considered as a unit of analysis by focusing brand-related sentiments (Sbr) {MLowSbr = 5.16, MHighSbr = 5.96; F (1,
on words and phrases used by the respondents (Aston et al., 2014). 343) = 30.98, p < .01} are found to be significant predictors of con-
Sentiment analysis was conducted using the Semantria® application sumers’ sharing intentions. Hence, sentiment scores (Sad and Sbr) were
(freely available on www.lexalytics.com/semantria/excel), which uses divided into “high” versus “low” categories by a median split and then a
a cloud-based corpus of words tagged with sentiments to analyze the two-step cluster analysis was conducted (Sarstedt and Mooi, 2014; Punj
dataset and then tag each sentence with a numerical sentiment score and Stewart, 1983) using these categorical variables, which helped
(Kim et al., 2016). This score ranges from −1.5 to +1.5 and the po- identify the most interpretable four-cluster solution in terms of practical
larity is categorised as (i) negative (ii) neutral or (iii) positive. relevance. This two-step method allows in discriminating natural
Besides demographic variables (gender and age), usage of Social groups from a set of variables stabilizing the nearness criterion, with a
Network Sites (SNS) was measured by asking the respondents the hierarchical agglomerative clustering, whose centers are far apart
question that how many hours do they spend online in a typical day for (Sarstedt and Mooi, 2014; Aldenderfer and Blashfield, 1984).
various activities: (i) SNS platforms like Facebook and Twitter, (ii) for In order to check for the relative and internal validity of cluster
study or work and (iii) total time spent on online activities. The cate- solution, Average Silhouette Coefficient Width (Rousseeuw, 1998;
gories of choices were: (1) 0 h; (2) 1–3 h; (3) 4–6 h; (4) 6–8 h, (5) 9–10 h Sarstedt and Mooi, 2014) and Pseudo F statistic (Caliński and Harabasz,
and (6) 11 h and more (Zhong et al., 2011). 1974; Sarstedt and Mooi, 2014) were evaluated by varying the number
of clusters. Four-cluster solution provided optimum goodness-of-fit
4. Results where the Average Silhouette Coefficient (ASC) was better (ASC = 0.7)
when compared to the two-cluster solution (ASC = 0.5) and three-
4.1. TLM versus sentiment analysis cluster solution (ASC = 0.6); also, there was no improvement in the
ASC for the five-cluster solution (ASC = 0.7). Additionally, the Pseudo
An ANOVA test revealed no significant interaction between ad ap- F statistic that describes the ratio of between-cluster variance to within-
peal and ad-related sentiments {F (1, 343) = 1.212, p > .10, ω2 cluster variance was found to be large (Pseudo F = 18.24) for the four-
= 0.004}. Hence, both the ads were entered simultaneously for the cluster solution, confirming the presence of close-knit and separated
hierarchical regression to assess the effect of cognitions/sentiments (ad- clusters. Four-cluster solution was also found to be reliable as the split-
and brand-related) and attitudes (Aad, Abr) on sharing intentions (SI). half cluster analysis validated the same (with ASC = 0.7) (Sarstedt and
TLM-based indicators of cognitive responses were entered in Model I Mooi, 2014; Punj and Stewart, 1983).
and sentiment-based indicators representing quantified cognitions were
entered in Model II. The results are presented in Table 1 below. 5. Findings
The results show that TLM-based indicators of cognitions (Cad, Cbr)
explained only 2.7% (β = 0.128, p < .05) variance in Step 1, whereas 5.1. Composition of clusters
the sentiment scores (Sad, Sbr) explained 9% (β = 0.285, p < .01) of
SI. In Step 2, attitudes (Aad and Abr) were entered and found to be Four clusters emerged, with sample sizes of 104 (30.2%), 103

4
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

Table 1
Results of the Hierarchical Regression Analysis.
Model I Model II
Thought-listing method Sentiment analysis
(Cad, Cbr) (Sad, Sbr)

Predictor Change in R2 F Change β Change in R2 F Change β

Step 1 Cognitions 0.027 4.682 0.128* 0.090 16.794 0.285**

Step 2 Attitude (Aad, Abr) 0.124 24.658 0.308* 0.109 23.031 0.279**
Final Model Adj. R2 = 0.140; F = 14.99* Adj. R2 = 0.189; F = 20.99**

Aad - Attitude towards ad content, Abr - Attitude towards embedded brand; Cad - Ad-related cognitions, Cbr - Brand-related cognitions; Sad - Ad-related sentiment
score, Sbr - Brand-related sentiment score; Significance at: *p < .05; * *p < .01, n = 344.

5.2. Cluster profiles

5.2.1. Active sharers (Cluster A)

This is the smallest cluster (17.2%) of SNS users, who are most likely
to share viral ads (SI = 6.55). Active sharers experience the same level
of intensity of affect for the ad as well as the brand, which is reflected in
their ad- and brand-related thoughts. Those who saw the Google ad,
describe it with words like ‘Good’, ‘touching’, ‘friendship’, ‘amazing’,
‘attractive’ and ‘impressive’; and they also think the brand Google to be
the ‘best’ and is the ‘top’ brand associated with their Internet ‘search’
and the brand is ‘innovative’ and a ‘useful’ means of ‘communication’.
Similarly, those who saw the ‘Volvo’ viral ad, described the ad execu-
tion with words such as ‘good’, ‘amazing’, ‘stability’, ‘music’, ‘cinema-
Fig. 2. Sentiment-based ABCD typology of viral ad sharers. tography’, ‘incredible’ and ‘unbelievable’ for the ‘brilliant’ and ‘creative’
‘stunt’. According to them, the brand ‘Volvo’ stands for ‘good’, ‘attrac-
(29.9%), 78 (22.7%) and 59 (17.2%) respondents, respectively. With tive’, ‘famous’ and ‘safety’, with its ‘international’ presence focused on
respect to sharing intentions (SI), the ANOVA test showed significant ‘innovating’ constantly in areas of ‘technology’ and ‘precision en-
mean differences between these four clusters {F (3343) = 18.24, gineering’ products, ‘worldwide’. Such thoughts reveal the strong sen-
p < .001}. Based on the relative importance that these ad sharers timents that the ‘active sharers’ have for these brands and they love
bestow on the ad- and brand-related sentiments, these clusters were forwarding emotionally-charged viral ads to their friends online. This
labeled as ‘Active sharers’, ‘Brand-fanatic sharers’, ‘Content-hungry cluster represents the ‘ideal seeds’ given the fact that they provide high
sharers’ and ‘Dormant sharers’ (See Fig. 2). We name this typology probability of fastening the diffusion of viral ads on social media plat-
using the acronyms of these four clusters, calling it the ‘ABCD’ typology forms.
of viral ad sharers.
Post hoc comparisons using the t-Test with Bonferroni correction 5.2.2. Brand-fanatic sharers (Cluster B)
further revealed the differences among four clusters for mean scores of This is the largest cluster (30.2%) of SNS users who are more tra-
ad- and brand-related sentiments (Range = −1.5 to +1.5). With re- ditional, displaying more affective gravity to the embedded brand vis-à-
spect to the ad-related sentiments (Sad), active sharers (MSad = 0.581, vis the ad content. For Brand-fanatics, a viral ad is perhaps perceived as
SD = 0.111) and content-hungry sharers (MSad = 0.584, SD = 0.117) being more effective or social if it is associated with portraying the
were found to be similar, as both these groups reported higher ad-re- functional benefits of the brand and uses linked with the embedded
lated sentiments. However, both active and content-hungry sharers brand. The word cloud reveals the use of more affective words to de-
were found to be significantly different from brand-fanatics (MSad = scribe the two brands (like ‘important’, ‘amazing’, ‘helpful’, ‘connecting’
0.143, SD = 0.214) and dormant sharers (MSad = 0.143, SD = 0.230) and ‘best’ for Google; ‘luxurious’, ‘reliable’, ‘safe’ and ‘trustworthy’ for
that represent low scores on ad-related sentiments. With regards to the Volvo) vis-à-vis the ads. Brand-fanatics think more about their con-
brand-related sentiments (Sbr), active sharers (MSbr = 0.567, SD = nection with the embedded brand within the viral ad, when it comes to
0.100) and brand-fanatics (MSbr = 0.569, SD = 0.147) were found to deciding whether to forward a particular ad or not. Their sharing in-
be similar as both of these groups reported higher brand-related sen- tention (SI = 5.38) is lower than that of the ‘Active sharers’.
timent score. Furthermore, both active and brand-fanatics sharers were
found to be significantly different from content-hungry (MSbr = 0.147, 5.2.3. Content-hungry sharers (Cluster C)
SD = 0.183) and dormant sharers (MSbr = 0.145, SD = 0.189) that This segment comprises of SNS users (29.9%) who experience
represent low scores on ad-related sentiments. Overall, active sharers stronger sentiments towards the ad ‘content’, regardless of the em-
(MSad = 0.581, MSbr = 0.567) represent the cluster of individuals bedded brand. They exhibit strong positive sentiments for the ad con-
carrying strong sentiments for both ad content as well as embedded tent and might discount any brand-related information if the ad content
brand. is sufficient to charge them emotionally. Word clouds of these ‘content-
For these four clusters, qualitative word maps were created using a hungry’ ad sharers highlight more affective words for the advertising
word cloud generator (www.wordclouds.com) based on participants’ message elements (e.g. ‘friends’, ‘childhood’, ‘partition’ and ‘relation-
ad- and brand-related cognitive responses, with the size of the word ship’ for describing the Google ad and ‘creative’, ‘interesting’, ‘sur-
reflecting the relative frequency of occurrence (See Table 2). Each of prising’ and ‘suspense’ to describe the Volvo ad) rather than the brand.
these types is discussed below: Content-hungry sharers love the emotional and provocative elements of

5
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

Table 2
Qualitative word maps for identified clusters of viral ad sharers.

viral ads, and are more likely to share such ads with their social net- male respondents, clusters differ in terms of the number of males in-
work connections. They score slightly higher on the variable of sharing cluded, with clusters A and C having more males than clusters B and D.
intention (SI = 5.40), when compared to brand-fanatics. Furthermore, results of the Kruskal-Wallis test indicate significant dif-
ferences across the four clusters in terms of total time individuals spend
5.2.4. Dormant sharers (Cluster D) for various online activities (χ2 = 7.537, df = 3, p < .10), but not
This segment, with the sample size of 22.7%, represents individuals with respect to the time spend on using SNS platforms (e.g. Facebook)
with weaker sentiments for both, the message as well as the embedded and using internet for working or studying purposes (p > .10).
brand within the viral ads, resulting in lowest sharing intentions (SI = As one of the viral ad, i.e. Google Search's Reunion ad, had the
4.93). These SNS users, who are ‘passive’ processors of viral content, context of India-Pakistan, the above results may be questioned on their
describing only facts and storylines from the ad (e.g. use of words like reliability, as the respondents are from India only. Also, the other ad,
‘Internet’, ‘people’, ‘search’ and ‘meet’ for describing ad-related Volvo, has a popular celebrity presence, which may have influenced the
thoughts in case of Google ad) without any affective feelings. Moreover, sharing intentions for this ad (Southgate et al., 2010). To validate the
they might be suspicious about the way the ad has been designed. For reproducibility of the proposed typology, another follow-up experiment
example, ‘Dormant’ respondents who saw the Volvo ad described it (n = 37) was conducted using another viral ad (https://youtu.be/
with words like ‘fictitious’, ‘confusing’, ‘photo-shopped’ and so on. uaWA2GbcnJU) of a lesser known brand (Thai Life Insurance) from
Hence, they must not be considered as a seeds to begin viral ad cam- the same pool of selected Top 10 viral ads, having no celebrity en-
paigns. dorser. The results reproduced a highly significant {F (3, 36) = 30.61,
p < .001} four-cluster solution just as in the main study, highlighting
5.3. External validity of the clusters the replicability of the proposed typology.

Criterion-related validity of the cluster solution was assessed using 6. Discussion and implications
demographic variables (gender and age) and SNS usage behaviour. Age
does not play a role in discriminating these four clusters. We found Marketing practitioners have largely benefited from various typol-
significant differences across clusters in terms of gender distribution (χ2 ogies proposed to identify different groups of Social Networking Site
= 10.373, df = 3, p < .05). Although the sample is biased towards users (Brandtzaeg and Heim, 2011), online shoppers (Bressolles et al.,

6
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

2014) as well as Facebook Fans (Wallace et al., 2014). Such structured about goods inside virtual communities and is seemed to be the most
segmentation of consumers help advertisers and brand managers similar to the typology proposed in our study. For example, in the
identify potential consumers easily and enhance the marketer's ability context of online community, Kozinet (1999) identifies four types of
to target important ‘seeds’ precisely. However, there was a lack of such users as – Tourists, Minglers, Devotees and Insiders. Tourists are users
structured typological framework to segment viral ad sharers. This who simply drop by the community every now and then with only
study adds to the existing literature by exploring a consumer typology superficial interest and few social ties. Minglers are users who maintain
based on cognitive responses as a psychographic variable. This type of strong social ties, while being marginally interested in any consumption
cognitive segmentation helps to group individuals based on their cog- activity. Devotees are users who maintain a strong interest in con-
nitive content and structure by capturing cognitions of each customer sumption but have little social attachments. Insiders are users who have
while incorporating their semantic uniqueness (Carrillat et al., 2009). strong social ties and a strong interest in consumption activity. On
This proposed sentiment-based ‘ABCD’ typology of viral ad sharers mapping these segments to that of the four clusters of viral ad sharers,
offers a comprehensive explanation of consumers’ sharing intentions we recognize that dormant sharers are like tourists, who hardly have
and has implications for targeting and content marketing strategies. any interest in either the advertising content or have any strong con-
The present study uses consumers’ cognitive responses and measures nections with the brand embedded in the viral ad. Content-hungry
them using sentiment analysis to propose four types of exclusive pro- sharers are minglers, who engage easily with the content but with
files of viral ad sharers. Each profile represents a unique combination of marginal interest or attachment with the embedded brand; whereas,
the varying strength of ad- and brand-related consumer sentiments and brand-fanatics are the devotees, who have strong attachment for the
exhibit significant differences in attitudes and behaviours. For example, embedded brand in the viral ad with relatively less interest in the ad
consumers who harbour both, strong ad- and brand-sentiments are most content. Finally, active sharers mimic insiders, who have equally strong
likely to pass-on viral ads. These hard to find ‘active’ consumers exist in connection with the brand and high interest in the ad content.
small proportion (only 17.2% of sample in this study) and make-up the Research is particularly warranted in understanding the meaning
pool of seeds that drive the diffusion of viral ads over social media behind commenting and sharing behaviour of SNS users (Dwivedi et al.,
platforms. Both online and offline research has theorized the existence 2015; Barger et al., 2016), in the context of viral advertising. To address
of a small segment of influential individuals and these are termed as this issue, an experimental set-up allowed this study to offer a holistic
innovators, who further influence the imitators (Bass, 1969; Hinz et al., view, combining instantaneous measurement of sharing intentions,
2011). For example, consistent with the classical Pareto principle (the along with attitudes (Aad and Abr) and cognitive responses. This pro-
law of the vital few), 20% of users are expected to carry 80% of the load vides more insights about the motives and beliefs of consumers that
to propagate the message. Therefore, it is crucial to wisely select the drive the intentions to forward viral ads. Another finding and con-
initial hosts for starting the epidemic viral phenomena. There is a tribution relates to the use of sentiment analysis to overcome the ap-
possibility that companies may fail in trying to create a viral marketing parently subjective nature of the traditional thought-listing method. By
epidemic because they spread the initial message too broadly (Kaplan using sentiment analysis, this research responded to an important call
and Haenlein, 2011). This study validates this finding that ‘active by Lutz and Swasy (1977) to develop deeper and objectively measure
sharers’ appear to be the smallest cluster in size and must be chosen as cognitive responses in terms of their valence as well as the strength.
an initial set of consumers to seed the viral campaign. Altogether, this study is a pioneering attempt to segment viral ad
Extant literature has demonstrated the importance of ‘message sharers based on their cognitive responses and also, makes an important
content’ as a single largest predictor of virality (Phelps et al., 2004; methodological contribution by introducing sentiment analysis as an
Dobele et al., 2007; Berger and Milkman, 2012) along with other pre- alternate method to overcome the limitations associated with the tra-
dictors like psychological motivations such as need to belong (e.g. Ho ditional thought-listing method.
and Dempsey, 2010) and features of social network structure such as tie
strength (e.g., Kiss and Bichler, 2008). Our proposed cluster profiles 6.1. Implications for managers
demonstrate that in addition to the message content (particularly
emotional content), positive sentiments towards ‘embedded brand’ in- The findings reported in this study have significant implications for
creases the probability of that content going viral. For example, ‘active advertisers and brand managers with respect to devising seeding stra-
sharers’ (i.e. strong sentiments for both the ad and the brand) have tegies as well as designing viral advertising campaigns. This study de-
higher intentions to share that ad when compared to ‘content-hungry’ monstrates that both, ad- and brand-related sentiments are strong dri-
(i.e. strong sentiments for ad only) or ‘brand-fanatic’ sharers (i.e. strong vers of sharing intentions. User generated content (UGC) from social
sentiments for brand only). Existing literature has also shown that high media platforms can provide richer understanding (Aswani et al.,
prominence of the ‘brand’ in an ad may distract consumers’ attention 2017c, 2018) about sentiments that a user carries for video content as
from content (Hsieh et al., 2012). Moreover, Huang et al. (2013) have well as the embedded brand. Brand managers can make use of other
shown that consumers’ are more likely to focus their attention on the freely available sentiment analysis tools like Python NLTK (Natural
content of a viral ad rather than the brand and if the ad has more brand Language Toolkit) and RapidMiner to analyze these comments and
information, it may lead to negative experience that reduces sharing identify active sharers as ‘seeds’ for their viral branding campaigns.
intentions. Contradictory to these studies, present study puts forth a The proposed typology of viral ad sharers will also help advertising
counter-intuitive finding that strong positive sentiments for a brand agencies in designing viral ads more strategically. In contrast to the
plays an important role along with more positive evaluations of the existing literature that advocates minimal brand information to be in-
message content; thereby, boosting consumers’ ad forwarding inten- cluded in an ad (Hsieh et al., 2012; Huang et al., 2013), this study
tions. This finding also supports the classical ‘Reciprocal Mediation’ highlights the necessity to balance the ad message with brand in-
model of advertising effectiveness proposed by MacKenzie et al. (1986), formation while producing viral ads, so as to prompt ‘active sharers’ at
which hypothesizes a reciprocal relationship between a consumer's ‘ad’ the initial stage of the campaign to boost virality. This may partly be
and ‘brand’ information processing behaviour. because strong brand integration is a sign of a well-structured video
The proposed meaningful categories of viral ad sharers can be (Southgate et al., 2010). In their study on exploring content virality on
mapped to the previous user typologies in the areas of adoption and Facebook, Aswani et al. (2017a) found that when a brand engages itself
consumption of products/information. Kozinets's (1999) typology is directly on the Facebook post via its name, then the probability of the
related to the consumption of products or the provision of information content going viral increases, indicating the ‘trust’ aspect that the brand

7
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

brings with itself to promote consumer engagement with the brand. when it comes to sharing of viral ads (Active, Brand-fanatics, Content-
Moreover, further including brand information supports consumers’ hungry and Dormant) and helps in easy identification of ‘seeds’, to
functional needs and draws their attention and motivates them to as- begin the viral campaign with and to design the viral message appro-
sociate the ad with a positive brand image (Lee and Hong, 2016). This priately for the selected segment.
finding supports the recent work of Akpinar and Berger (2017), which This study allows us to better understand the dynamics of in-
states that emotional ads where the brand is integrated into the nar- formation processing that leads to the formation of sharing intentions of
rative, boosts shares as well as the brand evaluation when compared to consumers when exposed to a viral advertisement by converting qua-
an emotional but non-integral ads and ads that are purely informative. litative cognitive responses into quantified data using sentiment ana-
For example, an advertisement for a cupcake brand in Pakistan, Peek lysis. Compared to the traditional thought-listing method, there are
Freans Cake Up recently garnered more than 10 million views on Fa- advantages of using social media analytics tools like sentiment analysis,
cebook with over 2,30,000 reactions, 63,000 shares and 7900 com- as this tool can analyze user-generated consumer responses in a con-
ments, shows a story of a working mother who finds a unique way to sistent manner, without the biases of the researchers. The findings from
create a real connection with her son using the product (Pattanaik, this study put forth positive brand evaluations as an equally important
2018). This ad integrates the brand very well in the storyline and po- predictor of consumers’ sharing intentions, along with message eva-
sitions itself in the consumer's mind as being bigger than just a sweet luation. Hence, marketers must make an attempt to design viral ads that
snack by showing how it can become the catalyst for starting real are not just emotional or entertaining but also convey what exactly the
conversations with the people we love. Hence, marketers must design brand stands for. Further, the results of this study suggest that the ef-
viral ads in such a way that they generate positive sentiments in the forts of practitioners looking to build successful viral advertising online
consumer's memory for the ad as well as brand. By doing so, they will should focus on engaging with the relatively small number of ‘active
be able to enthuse Active sharers who are motivated by both content as sharers’ who act as linchpins for the diffusion of the viral ad.
well as brand, thus, increasing the chances of the ad getting forwarded
by these active SNS users. Moreover, as demonstrated in the recent 7.1. Limitations and directions for future research
Twitter-based study on analysing polarization and acculturation in the
outcomes of the 2016 US Election campaign (Grover et al., 2018), First, the videos used in the experiments were limited in variety and
brand managers can identify the most impactful keywords, topics or were known to be viral. Each of the viral videos, one emotional and one
hashtags reported by active sharers. By using these cues, more social rational, was selected. An emotional viral video may have further sub-
media users can identify the link or location (Rathore et al., 2017) of dimensions like valence (positive vs. negative) and arousal levels (high
the viral campaign, where they can interact and discuss about the vs. low), which this research does not measure or control for. Future
particular ad or brand and eventually form the community of con- research may conduct field experiment or survey in order to measure
sumers that are most receptive to the brand's messages. For example, the actual sharing (rather than intention to share) and test more than
Ariel's #ShareTheLoad ad campaign went viral in 2015, where the two viral videos featuring messages varying in terms of valence and
brand managed to create a social media buzz using relevant hashtags arousal, as well as those ads that are not known to be viral. Second,
like #IsLaundryOnlyAWomansJob to promote gender equality. The gender imbalance is a limitation as 72% of the sample was males.
campaign garnered more than 1.5 Billion impressions and also led the However, in India, female constitute only 35% of the mobile internet
conversations where more than 1.5 Million men pledge to share the user base and only 24% of Facebook users are women; whereas, men
load (ET Brand Equity, 2016). constitute an overwhelming 76% of the user base (We Are Social,
Together, Brand-fanatics and Content-hungry ad sharers accounted 2016). Moreover, men spend over an hour per week on YouTube;
for about 60% of sample. They are not much different when it comes to whereas, women spend around 35 min per week consuming videos
sharing intentions. However, they are conceptually different; while the (Vermeren, 2015). So, the gender distribution in this study reflects the
former has high sentiments associated with the brand (but not the ad), actual gender distribution of the internet users in India. However, had
the latter has high sentiments for the content of the ad (but not with the the number of women been the same as men, we may have seen the
brand). Attention must be given to their communication preferences shift in the gender domination in some of the clusters. Finally, the viral
while designing viral ads. Finally, cluster D symbolizes Dormant videos used in this study involve different product categories and dif-
sharers, which represents lethargic consumers having least intentions to ferent levels of product involvement. Future researchers may validate
forward viral ads. Brand managers must develop a clear seeding these findings where these variables are controlled for.
strategy from the get-go. If they really do not know which user segment Research on sentiment-based segmentation of SNS users could
to target, they may want to consider testing one or more of above benefit from an investigation of different content consumption situa-
segments before singling one out. Overall, the above findings have tions. In particular, researchers are encouraged to study segmentation
practical implications for marketers, particularly in terms of segmen- of Internet users for different kinds of online communications like ads
tation and targeting, as well as for designing of viral messages. on Twitter, banner ads, movie teasers, and landing pages. Furthermore,
research may investigate on how other motivating factors like person-
7. Conclusion ality traits of SNS users and video characteristics (length, context,
creativity) might affect the valence or intensity of emotions and the
This study supports the notion that the user generated content in the feelings experienced by SNS users. This will help make better conclu-
form of cognitive responses carries rich information about consumers’ sions regarding the stability of the identified clusters overtime. Hence,
processing of viral advertisements that explains how the message as additional research is required to validate this consumer typology in
well as brand evaluation, together, drives individual's ad sharing in- other populations, along with additional profiling variables. Brands
tention. This study contributes to the field of viral advertising in two have a sizeable number of followers across social media platforms, yet
key areas: (1) using sentiment analysis, this study effectively measures they must be aware of the types of consumer segments that are present
cognitive responses of consumers, generated in response to viral ad- within these followers. In conclusion, this research is a pioneering at-
vertisements that capture valence as well as the strength of the cogni- tempt to identify and analyze the different types of viral ad sharers
tive responses; (2) using sentiment-based measures of ad- and brand- present on social media and also provides a working framework for
related cognitive responses, this study helps identify distinct consumer advertisers and brand managers to design and launch branded viral
segments that on the surface may look overlapping with each other campaigns.

8
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

Appendix A

(See Table A1)

Table A1
Viral advertisements used in the study.
1 URL: https://www.youtube.com/watch?feature=player_embedded&v=gHGDN9-oFJE
Google Search: Reunion Brand Name: Google
Upload Date: November 13, 2013
No. of Views: 12,259,717
Total Shares: 969,419
Shares on Facebook: 948,793
Shares on Twitter: 20,506
Shares on Blog Posts: 120

2 URL: https://www.youtube.com/watch?feature=player_embedded&v=M7FIvfx5J10
Volvo Trucks - The Epic Brand Name: Volvo
Split Upload Date: November 14, 2013
No. of Views: 78,605,590
Total Shares: 3,175,433
Shares on Facebook: 3,093,007
Shares on Twitter: 81,872
Shares on Blog Posts: 554

3 URL: https://www.youtube.com/watch?feature=player_embedded&v=uaWA2GbcnJU
"Unsung Hero" (Official Brand Name: Thai Life Insurance
HD) : TVC Thai Life Upload Date: April 9, 2014
No. of Views: 22,190,654
Insurance 2014 Total Shares: 1,191,139
Shares on Facebook: 1,160,751
Shares on Twitter: 30,029
Shares on Blog Posts: 359

#Note: Views and Share counts as of November 2014.

References review and research agenda. J. Res. Interact. Mark. 10 (4), 268–287.
Bass, F.M., 1969. A new product growth for model consumer durables. Manag. Sci. 15 (5),
215–227.
Akpinar, E., Berger, J., 2017. Valuable virality. J. Mark. Res. 54 (2), 318–330. Berger, J., Iyengar, R., 2013. Communication channels and word of mouth: how the
Alalwan, A., Rana, N.P., Dwivedi, Y.K., Algharabat, R., 2017. Social media in marketing: a medium shapes the message. J. Consum. Res. 40 (3), 567–579.
review and analysis of the existing literature. Telemat. Inform. 34 (7), 1177–1190. Berger, J., Milkman, K.L., 2012. What makes online content viral? J. Mark. Res. 49 (2),
Alalwan, A.A., 2018. Investigating the impact of social media advertising features on 192–205.
customer purchase intention. Int. J. Inf. Manag. 42, 65–77. Berger, J., Schwartz, E.M., 2011. What drives immediate and ongoing word of mouth? J.
Aldenderfer, M.S., Blashfield, R.K., 1984. Cluster Analysis. Sage, Beverly Hills, CA. Mark. Res. 48 (5), 869–880.
Aston, N., Liddle, J., Hu, W., 2014. Twitter sentiment in data streams with perception. J. Beverland, M., Dobele, A., Farrelly, F., 2015. The viral marketing metaphor explored
Comput. Commun. 2 (3), 11–16. through Vegemite. Mark. Intell. Plan. 33 (5), 656–674.
Aswani, R., Kar, A.K., Aggarwal, S., Ilavarsan, P.V., 2017a, November. Exploring Content Brandtzaeg, P.B., Heim, J., 2011. A typology of social networking sites users. Int. J. Web
Virality in Facebook: A Semantic Based Approach. Conference on e-Business, e- Based Communities 7 (1), 28–51.
Services and e-Society. Springer, Cham, pp. 209–220. Bressolles, G., Durrieu, F., Senecal, S., 2014. A consumer typology based on e-service
Aswani, R., Ghrera, S.P., Kar, A.K., Chandra, S., 2017b. Identifying buzz in social media: a quality and e-satisfaction. J. Retail. Consum. Serv. 21 (6), 889–896.
hybrid approach using artificial bee colony and k-nearest neighbours for outlier de- Cacioppo, J.T., Petty, R.E., 1981. Social psychological procedures for cognitive response
tection. Soc. Netw. Anal. Min. 7 (1), 38. assessment: the thought-listing technique. Cogn. Assess. 309–342.
Aswani, R., Chandra, S., Ghrera, S.P., Kar, A.K., 2017c, December. Identifying Popular Caliński, T., Harabasz, J., 1974. A dendrite method for cluster analysis. Commun. Stat.-
Online News: An Approach Using Chaotic Cuckoo Search Algorithm. In: Proceedings Theory Methods 3 (1), 1–27.
of the 2nd International Conference on Computational Systems and Information Carrillat, F.A., Riggle, R.J., Locander, W.B., Gebhardt, G.F., Lee, J.M., 2009. Cognitive
Technology for Sustainable Solution (CSITSS) . IEEE. pp. 1–6. segmentation: Modeling the structure and content of customers' thoughts. Psychol.
Aswani, R., Kar, A.K., Ilavarasan, P.V., Dwivedi, Y.K., 2018. Search engine marketing is Mark. 26 (6), 479–506.
not all gold: insights from Twitter and SEOClerks. Int. J. Inf. Manag. 38 (1), 107–116. Chen, T., Lee, H.M., 2014. Why do we share? The impact of viral videos dramatized to
Barger, V., Peltier, J.W., Schultz, D.E., 2016. Social media and consumer engagement: a sell. J. Advert. Res. 54 (3), 292–303.

9
K.K. Kulkarni, et al. Journal of Retailing and Consumer Services xxx (xxxx) xxxx

Chiu, H.C., Hsieh, Y.C., Kao, Y.H., Lee, M., 2007. The determinants of email receivers' Pattanaik S., 2018. This Pakistani ad about a mother and son will warm your heart,
disseminating behaviors on the Internet. J. Advert. Res. 47 (4), 524–534. retrieved from 〈https://www.hindustantimes.com/world-news/this-pakistani-ad-
Chu, S.C., 2011. Viral advertising in social media: participation in Facebook groups and about-a-mother-and-son-will-warm-your-heart/story-6pcZONYOGJlU2YcmT66jiO.
responses among college-aged users. J. Interact. Advert. 12 (1), 30–43. html〉 (Accessed 12 July 2018).
Correa, T., Hinsley, A.W., De Zuniga, H.G., 2010. Who interacts on the web?: the inter- Petrescu, M., Gironda, J., Korgaonkar, P.K., 2016. Hispanics and viral advertising. J.
section of users' personality and social media use. Comput. Hum. Behav. 26 (2), Retail. Consum. Serv. 32 (1), 46.
247–253. Petty, R.E., Cacioppo, J.T., 1979. Issue involvement can increase or decrease persuasion
Dobele, A., Lindgreen, A., Beverland, M., Vanhamme, J., Van Wijk, R., 2007. Why pass on by enhancing message-relevant cognitive responses. J. Personal. Social. Psychol. 37
viral messages? Because they connect emotionally. Bus. Horiz. 50 (4), 291–304. (10), 1915.
Dwivedi, Y.K., Kapoor, K.K., Chen, H., 2015. Social media marketing and advertising. Petty, R.E., Cacioppo, J.T., 1986. The elaboration likelihood model of persuasion. In:
Mark. Rev. 15 (3), 289–309. Berkowitz, L. (Ed.), Advances in Experimental Social Psychology 19. Academic Press,
Eckler, P., Bolls, P., 2011. Spreading the virus: emotional tone of viral advertising and its New York, pp. 123–205.
effect on forwarding intentions and attitudes. J. Interact. Advert. 11 (2), 1–11. Petty, R., Ostrom, T.M., Brock, T.C., 2014. Cognitive responses in persuasion. Psychology
Eslami, S.P., Ghasemaghaei, M., 2018. Effects of online review positiveness and review Press.
score inconsistency on sales: a comparison by product involvement. J. Retail. Phelps, J.E., Lewis, R., Mobilio, L., Perry, D., Raman, N., 2004. Viral marketing or elec-
Consum. Serv. 45, 74–80. tronic word-of-mouth advertising: examining consumer responses and motivations to
Greenwald, A.G., 1968. Cognitive learning, cognitive response to persuasion, and attitude pass along email. J. Advert. Res. 44 (4), 333–348.
change. Psychol. Found. Attitudes 147–170. Porter, L., Golan, G.J., 2006. From subservient chickens to brawny men: a comparison of
Grover, P., Kar, A.K., Dwivedi, Y.K., Janssen, M., 2018. Polarization and acculturation in viral advertising to television advertising. J. Interact. Advert. 6 (2), 30–38.
US election 2016 outcomes – can Twitter analytics predict changes in voting pre- Punj, G., Stewart, D.W., 1983. Cluster analysis in marketing research: review and sug-
ferences? Technol. Forecast. Soc. Change. https://doi.org/10.1016/j.techfore.2018. gestions for application. J. Mark. Res. 20, 134–148.
09.009. Rathore, A.K., Kar, A.K., Ilavarasan, P.V., 2017. Social media analytics: literature review
Hinz, O., Skiera, B., Barrot, C., Becker, J.U., 2011. Seeding strategies for viral marketing: and directions for future research. Decis. Anal. 14 (4), 229–249.
an empirical comparison. J. Mark. 75 (6), 55–71. Rousseeuw, P.J., 1998. Silhouettes: a graphical aid to the interpretation and validation of
Ho, J.Y., Dempsey, M., 2010. Viral marketing: motivations to forward online content. J. cluster analysis. J. Comput. Appl. Math. 20, 53–65.
Bus. Res. 63 (9), 1000–1006. Santos, R.L., Rocha, B.P., Rezende, C.G., Loureiro, A.A., 2007. Characterizing the
Hsieh, J.K., Hsieh, Y.C., Tang, Y.C., 2012. Exploring the disseminating behaviors of YouTube Video-sharing Community. Federal University of Minas Gerais (UFMG),
eWOM marketing: persuasion in online video. Electron. Commer. Res. 12 (2), Belo Horizonte, Brazil (Tech. Rep).
201–224. Sarstedt, M., Mooi, E., 2014. Cluster analysis. A Concise Guide to Market Research.
Huang, J., Su, S., Zhou, L., Liu, X., 2013. Attitude toward the viral ad: expanding tradi- Springer, Berlin, Heidelberg, pp. 273–324.
tional advertising models to interactive advertising. J. Interact. Mark. 27 (1), 36–46. Schulze, C., Schöler, L., Skiera, B., 2014. Not all fun and games: viral marketing for
Huang, Y., Hutchinson, J.W., 2008. Counting every thought: implicit measures of cog- utilitarian products. J. Mark. 78 (1), 1–19.
nitive responses to advertising. J. Consum. Res. 35 (1), 98–118. Shareef, M.A., Mukerji, B., Alryalat, M.A.A., Wright, A., Dwivedi, Y.K., 2018.
Kamboj, S., Sarmah, B., Gupta, S., Dwivedi, Y., 2018. Examining branding co-creation in Advertisements on Facebook: identifying the persuasive elements in the development
brand communities on social media: applying the paradigm of Stimulus-Organism- of positive attitudes in consumers. J. Retail. Consum. Serv. 43, 258–268.
Response. Int. J. Inf. Manag. 39, 169–185. Shareef, M.A., Mukerji, B., Dwivedi, Y.K., Rana, N.P., Islam, R., 2017. Social media
Kantar Millward Brown, 2018. Digital & Media Predictions 2018, Available from 〈http:// marketing: Comparative effect of advertisement sources. J. Retail. Consum. Serv. 46,
www.millwardbrown.com/Documents/MBArticles/Media_and_Digital_Predictions_ 58–69.
2018/?Access=yes〉 (Accessed 24 April 2018). Shiau, W.L., Dwivedi, Y.K., Yang, H.S., 2017. Co-citation and cluster analyses of extant
Kaplan, A.M., Haenlein, M., 2011. Two hearts in three-quarter time: how to waltz the literature on social networks. Int. J. Inf. Manag. 37 (5), 390–399.
social media/viral marketing dance. Bus. Horiz. 54 (3), 253–263. Shiau, W.L., Dwivedi, Y.K., Lai, H.H., 2018. Examining the core knowledge on Facebook.
Kapoor, K.K., Tamilmani, K., Rana, N.P., Patil, P., Dwivedi, Y.K., Nerur, S., 2018. Int. J. Inf. Manag. 43, 52–63.
Advances in social media research: past, present and future. Inf. Syst. Front. 20 (3), Sirgy, M.J., Grewal, D., Mangleburg, T.F., Park, J.O., Chon, K.S., Claiborne, C.B.,
531–558. Berkman, H., 1997. Assessing the predictive validity of two methods of measuring
Kim, J., Naylor, G., Sivadas, E., Sugumaran, V., 2016. The unrealized value of in- self-image congruence. J. Acad. Mark. Sci. 25 (3), 229–241.
centivized eWOM recommendations. Mark. Lett. 27 (3), 411–421. Southgate, D., Westoby, N., Page, G., 2010. Creative determinants of viral video viewing.
Kiss, C., Bichler, M., 2008. Identification of influencers—measuring influence in customer Int. J. Advert. 29 (3), 349–368.
networks. Decis. Support Syst. 46 (1), 233–253. Statista, 2017. Digital Advertising Report 2017 - Video Advertising, available from
Kozinets, R.V., 1999. E-tribalized marketing?: the strategic implications of virtual com- 〈https://www.statista.com/outlook/218/100/video-advertising/worldwide〉
munities of consumption. Eur. Manag. J. 17 (3), 252–264. (Accessed 12 May 2018).
Lee, J., Hong, I.B., 2016. Predicting positive user responses to social media advertising: Trusov, M., Bodapati, A.V., Bucklin, R.E., 2010. Determining influential users in Internet
the roles of emotional appeal, informativeness, and creativity. Int. J. Inf. Manag. 36 social networks. J. Mark. Res. 47 (4), 643–658.
(3), 360–373. Valos, M.J., Ewing, M.T., Powell, I.H., 2010. Practitioner prognostications on the future of
Libai, B., Muller, E., Peres, R., 2013. Decomposing the value of word-of-mouth seeding online marketing. J. Mark. Manag. 26 (3–4), 361–376.
programs: acceleration versus expansion. J. Mark. Res. 50 (2), 161–176. Vermeren, I., 2015, January. Men vs. women: Who is More Active on Social Media?,
Liu, B., 2010. Sentiment analysis and subjectivity. Handbook of Natural Language available from 〈https://www.brandwatch.com/blog/men-vs-women-active-social-
Processing 2. pp. 627–666. media/〉 (Accessed on 15 December 2017).
Liu, S.S., Stout, P.A., 1987. Effects of message modality and appeal on advertising ac- Wallace, E., Buil, I., de Chernatony, L., Hogan, M., 2014. Who “likes” you… and why? A
ceptance. Psychol. Mark. 4 (3), 167–187. typology of Facebook fans. J. Advert. Res. 54 (1), 92–109.
Liu-Thompkins, Y., 2012. Seeding viral content. J. Advert. Res. 52 (4), 465–478. We Are Social, 2016. Digital in APAC 2016. Available from 〈https://wearesocial.com/
Lovett, M.J., Peres, R., Shachar, R., 2013. On brands and word of mouth. J. Mark. Res. 50 special-reports/digital-in-apac-2016〉 (Accessed 20 November 2017).
(4), 427–444. Wright, P.L., 1973. The cognitive processes mediating acceptance of advertising. J. Mark.
Lutz, R., Swasy, J., 1977. Integrating cognitive structure and cognitive response ap- Res. 53–62.
proaches to measuring communication effect. In: Perreault Jr.William (Ed.), Wright, P.L., 1974. On the direct monitoring of cognitive response to advertising. In:
Advances in Consumer Research 4. Association for Consumer Research, Atlanta, GA, Hughes, G.D., Ray, M.L. (Eds.), Buyer/Consumer Information Processing. University
pp. 363–371. of North Carolina Press, Chapel Hill, NC, pp. 220–248.
MacKenzie, S.B., Lutz, R.J., Belch, G.E., 1986. The role of attitude toward the ad as a Www.ETBrandEquity.com, 2016, August 03. Ariel gets 2.1 million men to 'share the load'
mediator of advertising effectiveness: a test of competing explanations. J. Mark. Res. - ET BrandEquity. Retrieved December 30, 2018, from 〈https://brandequity.
130–143. economictimes.indiatimes.com/news/advertising/ariel-gets-2-1-million-men-to-
Misirlis, N., Vlachopoulou, M., 2018. Social media metrics and analytics in share-the-load/53517503〉.
marketing–S3M: a mapping literature review. Int. J. Inf. Manag. 38 (1), 270–276. Yun Yoo, C., Kim, K., 2005. Processing of animation in online banner advertising: the
Nelson-Field, K., Riebe, E., Newstead, K., 2013. The emotions that drive viral video. roles of cognitive and emotional responses. J. Interact. Mark. 19 (4), 18–34.
Australas. Mark. J. 21 (4), 205–211. Zajonc, R.B., Markus, H., 1982. Affective and cognitive factors in preferences. J. Consum.
Nisar, T.M., Prabhakar, G., Patil, P.P., 2018. Sports clubs' use of social media to increase Res. 9 (2), 123–131.
spectator interest. Int. J. Inf. Manag. 43, 188–195. Zhang, Y., Li, X., Wang, T.W., 2013. Identifying influencers in online social networks: the
Olson, J.C., Toy, D.R., Dover, P.A., 1982. Do cognitive responses mediate the effects of role of tie strength. Int. J. Intell. Inf. Technol. 9 (1), 1–20.
advertising content on cognitive structure? J. Consum. Res. 245–262. Zhong, B., Hardin, M., Sun, T., 2011. Less effortful thinking leads to more social net-
Pang, B., Lee, L., 2008. Opinion mining and sentiment analysis. Found. Trends Inf. Retr. 2 working? The associations between the use of social network sites and personality
(1–2), 1–135. traits. Comput. Hum. Behav. 27 (3), 1265–1271.

10