Conference PaperPDF Available

The Impact of Content, Context, and Creator on User Engagement in Social Media Marketing


Abstract and Figures

Social media has become an important tool in establishing relationships between companies and customers. However, creating effective content for social media marketing campaigns is a challenge, as companies have difficulty understanding what drives user engagement. One approach to addressing this challenge is to use analytics on user-generated social media content to understand the relationship between content features and user engagement. In this paper we report on a quantitative study that applies machine learning algorithms to extract textual and visual content features from Instagram posts, along with creator-and context-related variables, and to statistically model their influence on user engagement. Our findings can guide marketing and social media professionals in creating engaging content that communicates more effectively with their audiences.
Content may be subject to copyright.
The Impact of Content, Context, and Creator on
User Engagement in Social Media Marketing
Roope Jaakonmäki
University of Liechtenstein
Oliver Müller
IT University of Copenhagen
Jan vom Brocke
University of Liechtenstein
Social media has become an important tool in
establishing relationships between companies and
customers. However, creating effective content for
social media marketing campaigns is a challenge, as
companies have difficulty understanding what drives
user engagement. One approach to addressing this
challenge is to use analytics on user-generated social
media content to understand the relationship between
content features and user engagement. In this paper we
report on a quantitative study that applies machine
learning algorithms to extract textual and visual
content features from Instagram posts, along with
creator- and context-related variables, and to
statistically model their influence on user engagement.
Our findings can guide marketing and social media
professionals in creating engaging content that
communicates more effectively with their audiences.
1. Introduction
Over the past decade, social media has become a
popular channel through which to strengthen
customers’ relationships with products, brands, and
companies[20], [22], [27]. In a recent survey of 3,700
marketers, 96 percent of respondents answered that
they use social media for marketing [36].
However, as the number of end users and
marketers who are active on social media increases
[35], it becomes increasingly difficult for companies to
stand out from the crowd enough to engage their target
audiences. In fact, 91 percent of marketers struggle to
answer the question concerning the best ways to
engage their target audiences on social media platforms
[36]. What’s more, measuring the impact of social
media marketing campaigns is challenging, as is
calculating such campaigns’ return on investment [23].
In order to assess the success of social media
marketing activities, marketers typically measure the
rate at which users engage with their posts. The
engagement rate measures the quantity of responses
and interactions that content on social media generates
from users [4], [17], [31], [38]. How the engagement
rate is calculated varies across social media platforms,
but it generally measures the percentage of people who
react to a post in some way, such as by liking it or
commenting on it.
The factors that drive social media engagement can
be divided broadly into three groups: those that are
related to the post’s creator (e.g., the creator’s sex, age,
number of followers) [24], [21]; the post’s context
(e.g., time, location) [16], [41]; and certain features of
the content, such as, textual content (e.g., words, tags),
visual content (e.g., images, videos), and audio content.
While researchers have applied various methods to
study how users engage with textual content [2], [6],
[8], [9], [15], [21], [24], [26], [28], [34], [38], only a
few have focused on posts’ visual content [4], [5], [24].
Against this background, we follow a holistic
approach to study engagement in social media
marketing by statistically modeling the influence on
user engagement of the textual and visual features of
content on user engagement while controlling for
features related to creator and context. We use
machine-learning algorithms to extract the textual and
visual features of content from a dataset of more than
13,000 Instagram posts from professional bloggers and
to identify the most important features with regards to
user engagement. To the best of our knowledge, our
study is among the first to use a data-analytic approach
to identify automatically the most significant features
that drive social media engagement. Our results can
help social media marketers and users understand the
most effective approach to engaging social media
The remainder of this paper is structured as
follows. Section 2 gives a brief review of social media
marketing-related topics and summarizes existing
research on drivers of engagement on social media.
Section 3 explains the methodology and introduces the
dataset. Section 4 presents our empirical results, while
Section 5 discusses our findings, implications, and
limitations. Section 6 concludes with suggestions for
future research.
2. Background
2.1. Social media and influencer marketing
2.1.1. Social media marketing. In 2016 the number of
social network users reached 2.22 billion, a number
that is expected to increase to 2.72 billion by 2019
[35]. Because social media provides an inexpensive
way to interact and engage with these large numbers of
potential customers, social media marketing has
become a valuable channel for marketers [42]. The
purposes of using social media marketing include
branding, promotion, market research, customer
service, and customer relationship management
activities [11], [23], [42].
Social media marketing has three levels of maturity
[42]: trial, which includes testing various platforms but
not yet considering them an integral part of the
company’s marketing mix; transition, where social
media marketing activities are somewhat unplanned
but are becoming more systematic; and the strategic
phase, in which marketers have a formal process to
plan and execute social media marketing activities with
clearly defined objectives and metrics.
The effectiveness of social media marketing is
typically measured using proxies rather than
monetarily, as linking social media marketing activities
to key financial indicators is difficult [23]. Depending
on the goal, these proxy measures can include web
traffic generated, clicks, repeat visits, number of new
followers, search volume, mentions in other social
media channels, and peer-to-peer recommendations
[10]. Paine [31] suggests using engagement as a key
metric, dividing engagement into different phases,
starting with clicking and liking, continuing with
commenting, following, re-tweeting, and hash-tagging,
and finally evolving into advocacy.
2.1.2. Influencer marketing. Companies and
marketers use social media platforms not only to push
information about products to customers but also as a
medium for customer-to-customer communication
about product-related information, opinions, attitudes,
and purchase and post-purchase experiences [29]. In
fact, user-generated social media content has evolved
into a major factor in influencing consumer behavior
over the last years [23]. Therefore, it is not surprising
that marketing concepts like word-of-mouth (WOM)
and influencer marketing are gaining popularity among
social media marketers. WOM can be defined as the
act of consumers talking among themselves about a
product or service” [39, p. 280], while influencer
marketing can be seen as “the practice of identifying
key decision makers in a target audience and
encouraging them to use their influence to spread
WOM” [39, p. 277]. Thus, an influencer is a third party
who significantly shapes the opinions and purchasing
decisions of other customers [7]. For example,
influencers may post photos of themselves with
products or brands on a social media platform,
accompanied by brand-related hashtags, and be paid or
receive a free product from the brand in return as a
compensation. Influencers are often popular and well-
connected on social media. Although high popularity
and connectedness do not guarantee that a person has
significant influence and vice-versa [33], these
qualities are essential for influencers.
One of the main platforms for WOM and influencer
marketing is Instagram, which reached 400 million
users in 2015 [25]. According to a social media
marketing industry report, Instagram increased its
position most significantly among the top platforms
used by experienced social media marketers, increasing
from 28 percent in 2014 to 36 percent in 2015 [36].
Moreover, 52 percent of marketers are planning to
increase their Instagram marketing activities in the near
future [36]. Instagram data is also suitable for
analyzing the influence of content, since the posts
consist of both picture and text.
2.2. What drives social media engagement
In a survey of more than 1,500 marketers [32], 72
percent stated that their top social media priority is to
create more engaging content, and their second highest
priority (65%) is to improve their understanding of
what content is effective. These priorities are aligned
with our research aim to identify the factors that drive
engagement in social media marketing. Several
researchers have addressed particular aspects of this
question from a variety of perspectives, but holistic
research about what characterizes influential post is
still scarce. We divide the features that may influence
engagement into three categoriescreator, context,
and contentand elaborate in the following sections
on the current state of research in these areas.
2.2.1. Creator-related features. Many researchers
have studied creator-related features (e.g., the creator’s
number of followers, age, sex) for specific social
media communities. For example, Suh et al. [38] found
(not surprisingly) that the number of both followers
and followees affects the number of times a tweet is
retweeted on Twitter. Experience and age also
influence engagement. Arguello et al. [2] found, for
example, that posts on online communities were less
likely to get a reply if newcomers wrote them. The
same seems to hold for Twitter, as the age of a Twitter
account increases the number of retweets [38].
The gender of the account holder is another factor
that influences engagement on social media platforms.
Gilbert et al. [21], for example, discovered in their
study of Pinterest users that females get more repins
than men, although male Pinterest account holders
attract more followers than women do [21].
2.2.2. Contextual features. Most of the research done
on contextual features (e.g., time, location) has been
conducted and published by practitioners. For example,
TrackMaven analyzed the Instagram posts of 123
companies that are on the US Fortune 500 list [41] and
found that Sunday is the most effective day of the
week for posting and that the time of the day does not
have a significant effect on the number of interactions.
Similarly, Ellering gathered and analyzed sixteen
social media studies and found no best time to post
2.2.3. Content features. Content features can be
divided into the categories of text, visual, and audio
content. Several researchers have studied the textual
content’s effect on popularity. Berger and Milkman [6]
analyzed New York Times articles and found that
messages that include high-arousal positive emotions
(awe) and negative emotions (anger or anxiety) are
more likely to go viral than is content with other types
of emotions. Similarly, Lee et al. [28] observed that a
message that includes persuasive content (e.g.,
emotional and philanthropic) increases engagement,
while informative content (e.g., product prices,
availability, or features) reduces engagement when
used separately but increases engagement when
combined with some persuasive content.
Burke et al. [8] studied textual discourse in the
online community context and discovered that a short
group or topic introduction in messages increases
community response. In a later study, Burke and Kraut
[9] found that politeness increases the number of
replies in technical groups, but rudeness is more
effective in generating replies in political groups.
Arguello et al. [2] reported that posting on topic,
introducing oneself, asking questions, and using simple
language and shorter text increased replies in an online
Another prevailing trend in social media is the use
of hashtags and URLs. For example, hashtags and
URLs on Twitter have strong relationships with
retweets [38]. However, a study about Facebook
content found that the number of links in a post
decreases the number of comments [34]. According to
TrackMaven, posts by Fortune 500 companies that use
more than eleven hashtags provide the most
interactions [41].
Comparing social media posts with and without
visual content, Adobe found that posts with images
perform the best in engaging the audience in social
media [1]. Bakhshi, Shamma, and Gilbert [4]
performed probably the first study to look at which
visual features of a social media posts drive
engagement. Their research indicated that pictures that
include a human face are significantly more likely to
receive likes and comments than are photos without a
face. They also found that the number of faces in the
photo and the persons age and gender do not influence
engagement. In another study, Bakhshi et al. [5] found
that filtered photos attract more views and comments
than those without alterations. More specifically,
another study found that different Instagram filters
have different effects on the engagement rate [41].
Practitioners have generated other findings on
visual features. For instance, the social media
marketing analytics company Curalate [14] analyzed
eight million Instagram photos and discovered that
using light instead of dark images, blue as the
dominant color instead of red, duck-face selfies instead
of realistic selfies, low saturation instead of vibrant
colors, and a single dominant color instead of multiple
dominant colors generates more likes. Moreover,
Nielsen Norman Group found that users pay more
attention to photos with real people, big photos, and
images that carry information and tend to ignore
images that are too stimulating [30].
Finally, one widely acknowledged finding in the
marketing industry is that the gender and physical
attractiveness of a model in an ad seem to influence
people’s perception of the ad and the marketed product
[3]. One might assume that the same findings also
apply in the social media marketing context.
To the best of our knowledge, the audio features of
social media content have not yet been studied
3. Methods and data
We follow a quantitative approach to investigating
the relationship between creator-related, contextual,
and content features of Instagram posts and
engagement. While engagement can be quantified in
many ways, depending on the social media platform,
we measure engagement as the sum of likes and
comments. The number of likes indicates the extent of
interest and approval, and the number of comments
signals the level of verbal interaction, which also
signals user interest. In this section, we describe how
we collected and analyzed the data, and present a
statistical overview of our dataset.
3.1. Research process and data sampling
This study exploits an Instagram dataset from an
anonymous German marketing and advertising
company to determine which factors are most
influential on user engagement. Our research process
(Figure 1) started with collecting, processing, and
cleaning up the dataset, which consisted of a random
sample of Instagram posts. Then we gathered available
creator-related, contextual, and content variables from
the sample. Next, we filtered out the least common
variables to reduce the number of variables and created
a data frame for the regression. After filtering the
variables, we performed a least absolute shrinkage and
selection operator (LASSO) regression analysis on the
data frame to identify the most influential features.
Finally, we interpreted the results and compared them
with the findings from existing research.
Collecting a
dataset of 140 000
Instagram posts
Drawing a random
sample of 13 396
Instagram posts
Extracting creator-,
context-, and
features from the
Filtering out the
least common
Performing lasso
regression on the
Interpreting the
Figure 1. Research process
3.2. Extraction of features
We extracted creator- and context-related features
from the metadata provided by the Instagram API. The
creator-related data included variables like gender, age,
country of residence, number of followers, and number
of previous posts. Context-related features included the
time and date of the post.
To extract the content-related featuresthat is, the
text and visual featureswe wrote several Python
scripts using the Natural Language Tool Kit (NLTK)
and the Clarifai Image Recognition API.
For the textual features, we extracted the type and
number of words (including hashtags and URLs) and
emojis (a symbol expressing an emotion or an idea in
electronic messages) used in the posts’ caption fields.
In order to reduce the dimensionality of the resulting
feature vectors, we considered only those words and
emojis that appeared in at least 1 percent of the posts,
which resulted in 312 words and 114 emojis.
To capture the visual features, we used the Clarifai
image recognition API, which uses convolutional
neural networks to learn complex representations of
patterns in images [12], [43]. Clarifai’s API currently
consists of more than 11,000 classifiers, including
objects (e.g., car, house, river, man, woman), ideas
(e.g. education, love, leisure), and feelings (e.g.
beautiful, fun) [13]. The API is known for its accuracy,
which has been reported to be around 89.3 percent
[12]. Figure 2 shows a picture that illustrates the output
from classifying a picture with the Clarifai API. Using
the Clarifai API for the 13,396 Instagram photos
generated 2,061 unique classes, out of which we kept
only the 250 most frequent for the regression analysis.
Figure 2. Picture
and suggested
classifications according to Clarifai API: water,
woman, summer, travel, leisure, sea, relaxation,
vacation, young, enjoyment, recreation, fun,
ocean, one, beach, girl, tropical, outdoors
3.3. Regression analysis
After extracting the textual and visual content
features, along with the metadata for the creator and
context features, we created a data frame (13,396 rows
by 768 columns) to serve as input for the subsequent
regression analysis. Because of our dataset’s high level
of dimensionality, we chose to use LASSO regression,
which was first introduced by Tibshirani [40]. LASSO
is a linear regression method that performs variable
selection by shrinking the coefficients of uninfluential
independent variables to exactly zero, which produces
a model that includes only the most important
independent variables in explaining the dependent
variable [18], [40]. Model fitting was performed using
the “glmnet” package for R [19].
Although the LASSO technique is an advanced
regression method that works well with high numbers
of features [40], it has some limitations. In particular,
when a group of independent variables shows high
correlation, LASSO tends to pick one and set the
others to zero [18], [44], which may hinder the model’s
interpretability. We tried other variable-selection and
regularization approaches (e.g., elastic net and ridge
regression), but LASSO produced the best fit and
contained the lowest number of predictors.
3.4. Summary statistics of the dataset
Table 1 provides a statistical summary of some of
the independent variables of our regression model. In
particular, it presents the most often used words and
emojis and the most common image classes.
Table 1. Most common words, emojis, and image
1 875
2 570
5 539
1 469
2 028
4 988
1 075
2 018
4 497
1 268
3 424
1 218
3 399
1 178
2 583
1 141
2 394
1 076
2 242
2 154
2 124
Table 2 provides additional information about the
Instagram posts in our dataset. An average Instagram
post received 1,500 likes and thirty-five comments
(i.e., 50 times more likes than comments) and was
posted by a blogger with almost 60,000 followers. (Our
dataset stems from professional bloggers.) The caption
of the post averaged twenty-seven words and three
emojis, and the Clarifai API detected an average of
eight classes in the picture.
Table 2. Summary statistics of Instagram posts
1 546.04
58 185.40
Word count
Emoji count
Image classes
More than 80 percent of the posts were created by
female bloggers. The most common day to post was
Sunday, and the most common time to post was
between 8:00 p.m. and 9:00 p.m.
4. Results
The core functionality of LASSO regression is that
it can automatically perform variable selection and
explicate the tradeoff between highly accurate models
with many predictors and less accurate models with
fewer predictors. The plot in Figure 3 visualizes this
tradeoff. Each curve in the plot corresponds to one
predictor and the value of its coefficient (y-axis),
whereas the x-axis represents the amount of deviance
of the dependent variable that can be explained by the
predictors [19]. The plot shows, for example, that 40
percent of the deviance in engagement can be
explained by only 10 predictors, whereas increasing
the explanation to 50 percent% of the deviance
requires 381 predictors. The full model, with all 768
predictors, explains 51.39 percent of the deviance in
Figure 3. Amount of deviance explained by
Using the cross-validation functionality of the
glmnet package in R, we determined the optimal
tradeoff between the number of predictors and model
accuracy. The results (Figure 4) indicate that the
optimal number of variables lies around 383 variables,
so about half of the 768 features we extracted from the
Instagram posts have no significant influence on
Figure 4. Cross-validation curve with suggested
λ values
Table 3 presents the most influential creator- and
context-related predictors. The table shows the
frequency with which these variables occur in the
dataset, as well as their estimated regression
coefficients. The coefficients can be interpreted as the
additional number of likes and comments the post
would be predicted to receive if these features were
present. For example, a twenty-year-old woman
posting on Friday night at 8:00 p.m. is predicted to
receive 1,332 more likes and comments than the
average post.
Table 3. Most influential creator- and context-
related predictors
Posting time a.m.
6:00-7:00 a.m.
Posting time p.m.
8:00-9:00 p.m.
1 097
Posting day
1 925
10 997
2 192
Born before 1995
4 162
Similarly, Table 4 shows the most influential
content-related predictors. For example, a post that
includes the caption “Wonderful Switzerland, a
speak-no-evil monkey emoji, and a picture of a
woman on a mountain is predicted to receive 2,096
more likes and comments than average.
Table 4. Most influential content-related
predictors (i.e., words, emojis, and image classes)
1 011
Image class
1 175
5 539
5. Discussion, implications, and limitations
5.1. Contribution to practice
Our results indicate that choosing the right
influencer affects user engagement, as the creator-
related factorsespecially the number of followers
and the creator’s age and genderplay the most
significant role among all predictors. Similarly, there
are certain days and hours (i.e., contexts) during which
the audience is more likely to be engaged than at
others. Influencer marketing professionals can use this
information to choose bloggers and define the launch
time of social media marketing campaigns. Our
findings regarding content features can also guide
content-creation strategies for social media marketing,
thereby responding to marketers’ need to improve their
understanding of what types of content are the most
engaging [1], [36]. For example, our results suggest
that pictures with people and scenery and emojis that
express positive emotions (e.g., relief, love, joy)
increase engagement.
An open issue to be explored in future work
concerns the consequences of designing content
according to this knowledge. For example, it is
possible that, as professional bloggers and marketers
increase their use of content that is predicted to be
highly engaging, the content will lose effectiveness: If
everybody posts pictures with women in front of nature
scenes on Friday evenings at 8:00 p.m., a kind of
fatigue effect may set in. However, the approach we
presented here can easily be repeated with minimal
costs in order to monitor such developments in near-
real time.
5.2. Contribution to research
To the best of our knowledge, this study is the first
to include content variables like words, emojis, and
images as independent variables to explain engagement
on social media platforms. Even though none of the
content features in the social media posts alone
explains more than 2 percent of the deviance in
engagement, these features are easy to influence and
combine in order to increase a post’s impact.
The findings of our research also extend the
existing body of literature by confirming the
importance of contextual features (e.g., date, time) that
practitioners have identified [41] [16] and the high
impact of creator-related features (e.g., age, sex,
followers) [2], [21]. Of all the features we examined,
we found that the number of followers has the most
impact on user engagement [4]. Although this finding
is not so surprising, unlike context and content
features, this variable is difficult for users to influence.
As for the content factors that drive engagement, our
findings confirm that people in pictures increase the
engagement rate and that pictures that include text and
scenery have a high impact on the number of likes and
comments received [4], [30]. However, we speculate
that these content-related features are highly dependent
on the industry that uses them and on how they are
5.3. Limitations
One limitation of this study is the limited
generalizability of the results, as the dataset contains
only information from Instagram bloggers from
German-speaking countries. A similar approach to the
one presented here could easily be used with data from
other social media platforms to increase the results’
A methodological limitation associated with the
study is that we used LASSO regression, which selects
only one feature and sets the others to zero in a case
when features are highly correlated. Hence, our
analysis may have missed features that are highly
correlated with those presented in our Results section.
In addition, LASSO regression does not provide
information regarding the statistical significance of
predictors; we can trust it only to discard the
insignificant variables and select the significant ones. It
is also difficult to argue the predictive accuracy of our
model (51.39% of deviance explained), as there are no
comparable models reported in the literature to use as
Moreover, using an automated approach to classify
pictures sometimes results in misclassification, even
though manual checks of the Clarifai results indicated
a high level of accuracy.
In future studies, we intend to provide a more
comprehensive and precise view of what drives social
media engagement. Using the sum of likes and
comments might not be the best proxy for engagement
because the ratio and weight between comment and
likes is not balanced, so we will consider using other
measures for engagement.
Finally, as we used only a static snapshot of data in
our analysis, we were not able to capture fully the
dynamic nature of engagement on social media
platforms. Some of the posts we analyzed might have
received more likes and comments after we
downloaded the data, which might have caused biases
in our analysis.
6. Conclusion
Our approach to identifying and quantifying the
factors that influence engagement in social media
marketing demonstrates how data analytics can create
business value for marketing organizations. Besides
directly applying the insights we generated from our
analysis, the approach we used can be used in business
contexts to maximize the impact of social media
activities and increase interaction with potential
customers. For instance, our results and approach can
guide companies and influence marketers to create
more appealing advertisements and successful WOM
marketing campaigns by designing engaging content
and choosing influential creators and contexts.
This research also creates a foundation for future
research on social media engagement. For example,
future research may seek to identify additional features
that increase the ability to explain and predict
engagement and may study whether and how
predictors of engagement differ based on the use
contexts (e.g., different products, brands, or industries).
Finally, our approach might also be used to predict and
improve the impact of social media posts in
applications outside of marketing (e.g., politics).
7. References
[1] Adobe Digital Index, The Social Intelligence Report - Q1,
2014, available at:
(retrieved June 8, 2016).
[2] Arguello, J., B. S. Butler, E. Joyce, R. Kraut, K. S. Ling,
C. Rosé, and X. Wang, “Talk to Me: Foundations for
Successful Individual-Group Interactions in Online
Communities”, in: Proceedings of the SIGCHI Conference
on Human Factors in Computing Systems, ACM, Montreal,
Québec, 2006, pp. 959-968.
[3] Baker, M. J., and G. A. Churchill Jr, “The Impact of
Physically Attractive Models on Advertising Evaluations”,
Journal of Marketing Research, 14(4), 1977, pp. 538-555.
[4] Bakhshi, S., D. A. Shamma, and E. Gilbert, “Faces
Engage Us: Photos with Faces Attract More Likes and
Comments on Instagram”, in: Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems, ACM,
Toronto, ON, 2014, pp. 965-974.
[5] Bakhshi, S., D. A. Shamma, L. Kennedy, and E. Gilbert,
“Why We Filter Our Photos and How It Impacts
Engagement”, in: Proceedings of the Ninth International
Conference on Web and Social Media, AAAI, Oxford,
England, 2015, pp. 12-21.
[6] Berger, J., and K. L. Milkman, “What Makes Online
Content Viral?”, Journal of Marketing Research, 49(2), 2012,
pp. 192-205.
[7] Brown, D., and N. Hayes, Influencer Marketing: Who
Really Influences Your Customers?, Routledge, Amsterdam,
Netherlands, 2008.
[8] Burke, M., E. Joyce, T. Kim, V. Anand, and R. Kraut,
“Introductions and Requests: Rhetorical Strategies That Elicit
Response in Online Communities”, in: Proceedings of the
Third Communities and Technologies Conference, Springer
London, Michigan State University, MI, 2007, pp. 21-39.
[9] Burke, M., and R. Kraut, “Mind Your Ps and Qs: The
Impact of Politeness and Rudeness in Online Communities”,
in: Proceedings of the Conference on Computer-Supported
Cooperative Work, ACM, San Diego, CA, 2008, pp. 281-
[10] Castronovo, C., and L. Huang, “Social Media in an
Alternative Marketing Communication Model”, Journal of
Marketing Development and Competitiveness, 6(1), 2012,
pp. 117-131.
[11] Chen, Y., S. Fay, and Q. Wang, “The Role of Marketing
in Social Media: How Online Consumer Reviews Evolve”,
Journal of Interactive Marketing, 25(2), 2011, pp. 85-94.
[12] Clarifai, Technology, 2016, available at: (retrieved June 13,
[13] Clarifai, The ultimate guide to visual recognition APIs,
2016, available at:
visual-recognition-api-guide/#.V1r9Idl97IV (retrieved June
13, 2016).
[14] Curalate, Pixel perfect: Which image qualities drive
engagement on Instagram, 2013, available at:
more-instagram-likes/ (retrieved June 8, 2016).
[15] Cvijikj, I. P., and F. Michahelles, “A Case Study of the
Effects of Moderator Posts within a Facebook Brand Page”,
in: Proceedings of the Third International Conference on
Social Informatics, Springer Berlin Heidelberg, Singapore,
2011, pp. 161-170.
[16] Ellering, N., What 16 Studies Say About the Best Times
to Post on Social Media, 2016, available at:
media/ (retrieved June 8, 2016).
[17] Facebook, How is Engagement Rate Defined?, 2016,
available at:
(retrieved June 6, 2016).
[18] Friedman, J., T. Hastie, and R. Tibshirani,
“Regularization Paths for Generalized Linear Models via
Coordinate Descent”, Journal of Statistical Software, 33(1),
2010, pp. 1-22.
[19] Friedman, J., Hastie, T., and Tibshirani, R., Package
´glmnet´, 2016, available at: https://cran.r- (retrieved June
2, 2016).
[20] Gandomi, A., and M. Haider, “Beyond the Hype: Big
Data Concepts, Methods, and Analytics”, International
Journal of Information Management, 35(2), 2015, pp. 137-
[21] Gilbert, E., S. Bakhshi, S. Chang, and L. Terveen, “I
Need to Try This?: A Statistical Overview of Pinterest”, in:
Proceedings of the SIGCHI Conference on Human Factors in
Computing Systems, ACM, Paris, France, 2013, pp. 2427-
[22] He, W., S. Zha, and L. Li, “Social Media Competitive
Analysis and Text Mining: A Case Study in the Pizza
Industry”, International Journal of Information Management,
33(3), 2013, pp. 464-472.
[23] Hoffman, D. L., and M. Fodor, “Can You Measure the
ROI of Your Social Media Marketing?”, MIT Sloan
Management Review, 52(1), 2010 pp. 41-49.
[24] Hu, Y., L. Manikonda, and S. Kambhampati, “What We
Instagram: A First Analysis of Instagram Photo Content and
User Types”, in: Proceedings of the 8th International
Conference on Weblogs and Social Media, The AAAI Press,
Ann Harbor, US, 2014, pp. 595-598.
[25] Instagram, Celebrating a Community of 400 Million,
2015, available at:
400million (retrieved June 6, 2016).
[26] Jamali, S. and H. Rangwala, “Digging Digg: Comment
Mining, Popularity Prediction, and Social Network
Analysis”, in: Proceedings of the International Conference on
Web Information Systems and Mining, IEEE, Shanghai,
China, 2009, pp. 32-38.
[27] Laroche, M, M. R. Habibi, and M. O. Richard, “To Be
or Not to Be in Social Media: How Brand Loyalty is
Affected by Social Media?”, International Journal of
Information Management, 33(1), 2013, pp. 76-82.
[28] Lee, D., H. Kartik, and N. Harikesh, "The Effect of
Social Media Marketing Content on Consumer Engagement:
Evidence from Facebook", Working paper, The Wharton
School, University of Pennsylvania, 2014, pp. 1-51.
[29] Mangold, W. G., and D. J. Faulds, “Social Media: The
New Hybrid Element of the Promotion Mix”, Business
Horizons, 52(4), 2009, pp. 357-365.
[30] Nielsen, J., Photos as Web Content, 2010, available at:
(retrieved June 8, 2016).
[31] Paine, K. D., Measure What Matters: Online Tools for
Understanding Customers, Social Media, Engagement, and
Key Relationship, John Wiley & Sons, Westford, MA, US,
[32] Pulizzi, J., and Handley, A., B2B Content Marketing:
2016 Benchmarks, Budgets and Trends North America,
2015, available at:
marketing-research/ (retrieved June 8, 2016).
[33] Romero, D. M., W. Galuba, S. Asur, and B. A.
Huberman, “Influence and Passivity in Social Media”, in:
Proceedings of the European Conference on Machine
Learning and Knowledge Discovery in Databases, Springer
Berlin Heidelberg, Athens, Greece, 2011, pp. 18-33.
[34] Sabate, F., J. Berbegal-Mirabent, A. Cañabate, and P. R.
Lebherz, “Factors Influencing Popularity of Branded Content
in Facebook Fan Pages”, European Management Journal,
32(6), 2014, pp. 1001-1011.
[35] Statista, Number of Social Network Users Worldwide
from 2010 to 2019 (in billions), 2016, available at:
worldwide-social-network-users/ (retrieved June 8, 2016).
[36] Stelzner, M. A., Social Media Marketing Industry
Report: How Marketers Are Using Social Media to Grow
Their Businesses, 2015, available at:
gIndustryReport2015.pdf (retrieved June 8, 2016).
[37] Sterne, J., Social media metrics: How to Measure and
Optimize Your Marketing Investment, John Wiley & Sons,
Hoboken, NJ, US, 2010.
[38] Suh, B., L. Hong, P. Pirolli, and E. H. Chi, “Want to Be
Retweeted? Large Scale Analytics on Factors Impacting
Retweet in Twitter Network”, in: Proceedings of the Second
International Conference on Social Computing, IEEE,
Minneapolis, MN, 2010, pp. 177-184.
[39] Thorne, L., Word-of-mouth Advertising, Online and
Off: How to Spark Buzz, Excitement, and Free Publicity for
Your Business or Organization with Little or No Money,
Atlantic Publishing Company, Ocala, FL, 2008.
[40] Tibshirani, R., “Regression Shrinkage and Selection via
the Lasso”, Journal of the Royal Statistical Society: Series B
(Methodological), 1996, pp. 267-288.
[41] Trackmaven, The Fortune 500 Instagram Report, 2014,
available at:
une500_Instagram_Report.pdf (retrieved June 6, 2016).
[42] Tuten, T. L., and M. R Solomon, Social Media
Marketing, Sage, London, UK, 2014.
[43] Zeiler, M. D., and R. Fergus, “Visualizing and
Understanding Convolutional Networks”, in: Proceedings of
the European Conference on Computer Vision, Springer
International Publishing, Zürich, Switzerland, 2014, pp. 818-
[44] Zou, H., and T. Hastie, “Regularization and Variable
Selection via the Elastic Net”, Journal of the Royal Statistical
Society: Series B (Statistical Methodology), 67(2), 2005, pp.
... The authors proposed a model capable of classifying mentions into three categories based on its motivation: information-oriented, relationship-oriented, and discussion-oriented. Jaakonmäki et al. [121] analyzed the inŕuence of the content posted for social media marketing. They used machine learning algorithms to extract textual and visual content features from posts, along with creator and context features, to model their inŕuence on user engagement. ...
... Therefore, it is of utmost importance to understand how users interact with each other to őnd out how information is disseminated on the platform and how online debate affects our society. Prior studies of user behavior on Instagram have mainly focused on user engagement according to content type [243,121,91,133,134,311], general characteristics of comments related to political messages [292,336], and the impact of posted content on marketing contexts [121,324,132]. However, the literature lacks an investigation of the networks that emerge from users' interactions, particularly in the context of political content that fosters the spread of information. ...
... Therefore, it is of utmost importance to understand how users interact with each other to őnd out how information is disseminated on the platform and how online debate affects our society. Prior studies of user behavior on Instagram have mainly focused on user engagement according to content type [243,121,91,133,134,311], general characteristics of comments related to political messages [292,336], and the impact of posted content on marketing contexts [121,324,132]. However, the literature lacks an investigation of the networks that emerge from users' interactions, particularly in the context of political content that fosters the spread of information. ...
Full-text available
Understanding the collective behavior of (groups of) individuals in complex systems, even in scenarios where the individual properties of their components are known, is a challenge. From the point of view of network models, the collective actions of these individuals are often projected on a graph forming a network of co-interactions, which we here refer to as a many-to-many network. However, the volume and diversity with which these co-interactions are observed in the most varied systems, such as, for example, social media platforms, economic transactions and political behavior in voting systems, impose challenges in the extraction of patterns (structural, contextual and temporal) emerging from collective behavior and that are fundamentally related to a phenomenon under study. Specifically, the frequent presence of a large number of weak and sporadic co-interactions, which, therefore, do not necessarily reflect patterns related to the phenomenon of interest, end up introducing noise to the network model. The large amount of noise, in turn, may obfuscate the most fundamental behavior patterns captured by the network model, that is, the patterns that are essentially relevant to the understanding of the phenomenon under investigation. Removing such noise becomes then a key challenge. Our goal in this dissertation is to investigate the modeling and analysis of collective behavior patterns that emerge in networks formed by co-interactions in different contexts, aiming to extract relevant and fundamental information about a target phenomenon of interest. Specifically, we tackle the extraction of structural, contextual and temporal properties associated with patterns of collective behavior that are fundamentally represented by communities extracted from the network. To this end, we propose a general strategy that addresses the aforementioned challenges. In particular, this strategy includes, as an initial step, the identification and extraction of the network backbone, that is, the subset of the edges that are indeed relevant to the target study. The next steps consist of the extraction of communities from this backbone as a manifestation of the existing collective behavior patterns and the characterization of the structural (topological), contextual (related to the phenomenon of interest) and temporal (dynamic) properties of these communities. Based on this general strategy, we propose specific artifacts for some of the steps that compose it and advance the state-of-the-art, in particular with a new method for backbone extraction, a new temporal node embedding method capable of representing and extracting different temporal patterns of interest from a sequence of networks, and finally a methodology to support the selection and evaluation of backbones from a structural and contextual point of view, considering the most common scenario where there is no ground truth. Furthermore, we explore these artifacts by studying three different phenomena that require different modeling and analysis strategies. Specifically, we investigate: (i) the formation of ideological groups in the Brazilian and U.S. House of Representatives, (ii) online discussions on Instagram in Brazil and Italy, and (iii) information dissemination on WhatsApp. Overall, our results show that the proposed artifacts offer relevant contributions to the field in which this dissertation is inserted.
... There are three core factors that influence user actions on social media (Jaakonmäki et al., 2017). Whether the content generates many comments, shares, or receives "likes" or "dislikes" it is driven by, (1) user account (e.g., gender of content creator, account age etc.), (2) context of post, (e.g., referenced location, time of day), and (3) content-specific features of the post (e.g., function of post, extent of personification, are human faces in the content visible, voices in the audio etc.) (Jaakonmäki et al., 2017). ...
... There are three core factors that influence user actions on social media (Jaakonmäki et al., 2017). Whether the content generates many comments, shares, or receives "likes" or "dislikes" it is driven by, (1) user account (e.g., gender of content creator, account age etc.), (2) context of post, (e.g., referenced location, time of day), and (3) content-specific features of the post (e.g., function of post, extent of personification, are human faces in the content visible, voices in the audio etc.) (Jaakonmäki et al., 2017). Few studies have looked at the role of functions of posts and how they influence engagement actions (Kanol and Nat, 2021;Ischen et al., 2020). ...
Full-text available
The value experience perceived by users and the extent of interactivity on social media show how engaging audiences are. Few studies have looked at what drives this value experience in organizational communication. This study explores the functional use of communications by interest group organizations (IGOs) and discerns their effect on user engagement with and without multimedia inclusion on Twitter. A bi-term topic modeling technique is used to analyze posts from 121 organizations, and a generalized linear regression model to assess the link between the content functions and user engagement. The results show that the information and communication content functions include event updates and people recognition. Further, report, event, period, and people communication functions drive a higher engagement with multimedia inclusion, while unite, sign, and glean communication functions are more likely to increase engagement without multimedia elements. This study bridges the gap in the service literature as it pertains to non-profit organizations (i.e., interest group organizations) by exploring organizational communication using communications content functions of Twitter posts. This study is the only one to investigate content functions beyond the categorizations of message functions and the relationship between content functions and user engagement.
... A user might be an individual or organization, and content may take many forms. While some 97% of marketers state that they use SM for marketing, effective methodologies remain largely undefined and difficult to quantify (Jaakonmäki et al., 2017). The standard to measure the success of a marketing strategy, return on investment (ROI), appears to be a poor fit for SMM as ROI fails to evaluate direct customer interaction and the long-term benefits of relationships built between companies and customers (Hoffman and Fodor, 2010). ...
... Agencies often treat online content as traditional media, unidirectionally broadcasting messaging to stakeholders while ignoring engagement and relationship building (Kent, 2013). Building an online community of adequate size appears to be an additional hindrance to effective content marketing use (Jaakonmäki et al., 2017). However, through the use of IM, many of these shortfalls can potentially be overcome. ...
Research Proposal
Full-text available
From industrial contamination to the collapse of biological diversity, decision-makers and environmental managers are faced with increasing pressure to ensure the health and productivity of the environment. Creating effective and long-lasting solutions is anything but simple, and requires inputs from groups such as managers, scientists, and stakeholders. Until the last 20-years, the options available to environmental agencies to identify and engage stakeholders were largely reliant on news media, word-of-mouth, and government notifications. With the popularization of the internet and the growth of social media, a unique opportunity now exists for environmental agencies to engage with the human aspect of their efforts. Through the use of social media, individuals have self-organized around similar topics of interest. These groups are often diverse in ethnicity, and background, and can hold information key for effective management. In essence, stakeholders once decentralized, have coalesced online, providing virtual spaces environmental agencies can find, and engage with, stakeholders. Similar methodologies are regularly used in the private sector, called social media marketing, but research into the effectiveness in the context of government stakeholder engagement is limited. This project will test the efficacy of environmental managers’ use of social media marketing for stakeholder engagement. Through a partnership with the State of Hawaiʻi Department of Aquatic Resources (DAR), I will identify online populations of stakeholders, and engage them with the use of social media marketing. The effectiveness of this method will be determined through comparisons of event attendance data, cost analysis, and attendee questionnaires. I will summarize successful methodology and make it publicly available, allowing for further use and development of the strategy by additional researchers and environmental agencies. More effective stakeholder engagement methods will result in improved regulations and management practices, provide a sense of ownership to stakeholders, and enable creative solutions to environmental issues.
... It is intended that users share information of personal, inherent and intimate value in real time from their experiences, facilitating feedback and generating a repository of strategic knowledge and be a source of inspiration for travelers (Guerrero, Møller, Olafsson, & Snizek, 2016). On the other part, the rapid expansion, variety of interactions and content that characterize social networks, make it an important source of Big Data generation (Parady, et al., 2019;Jaakonmäki, Müller, & Vom Brocke, 2017;Oviedo, Muñoz, & Castellanos, 2015). The combination of Big Data and social networks creates value to realize smart tourism, improves decision making, enriches the tourist experience, develops new products and business models (Del Vecchio, Mele, Ndou, & Secundo, 2018). ...
Full-text available
Planning for the transformation of a destination into a smart tourist destination must consider the use of state-of-the-art technology as an essential requirement. This article describes how to use technology to drive the process of transforming Costa Rica's La Fortuna destination into a smart tourist destination. The methodology was assumed through a multimethod design in which a case study was developed in La Fortuna destination in collaboration with the Arenal Cámara de Turismo y Comercio. Therefore, it was possible to determine and characterize the technologies used in the main smart tourist destinations, the SWOT analysis factor of La Fortuna tourist destination with the greatest applicability of technologies and the design of an information system for La Fortuna. These results can be used as a reference point by other tourist destinations with similar characteristics to La Fortuna and that wish to start the process of transformation into a smart tourist destination. RESUMEN La planificación para la transformación de un destino en un destino turístico inteligente debe contemplar el uso de tecnología de punta como un requerimiento esencial. Este artículo describe, cómo emplear la tecnología para impulsar el proceso de transformación del destino La Fortuna de Costa Rica en un destino turístico inteligente. La metodología fue asumido a través de un diseño multimetódico en colaboración de la Arenal Cámara de Turismo y Comercio. Por lo que, se logró determinar las tecnologías utilizadas en los principales destinos turísticos inteligentes, el factor del análisis FODA del destino turístico La Fortuna con mayor aplicabilidad de tecnologías y el diseño de un sistema de información para La Fortuna. Estos resultados pueden ser utilizados como punto de referencia por otros destinos turísticos con características similares a La Fortuna y que desean iniciar el proceso de trasformación en un destino turístico
Discussing Turkey's cultural diplomacy devices around some popular culture content determines the tendencies of researchers and communication professionals. Cultural diplomacy relations, shaped as a field of ideological struggle and competition for reputation, also become the catalyst forces of a history of representation in which nations are staged as brand value in an age where fluid cultures converge. Although the idea of marketability of culture attracts the production movements of the cultural industries concept and critical theory, it is seen that the market created by cultural diplomacy has a unique industrial value. This study aims to discuss alternative trends for Turkey's cultural diplomacy activities that are reduced to some devices and are likely to enter a vicious circle, and to build the initial interest required to integrate these trends into an integrated marketing tool for controlled image actions. Because all the moves of the culture industries require trans-media narrative synergy as part of an integrated marketing strategy. In this context, Autonomous Sensory Meridian Response (ASMR), which is discussed as a cultural diplomacy device, is important because it is a new type of stimulative/triggering narrative style that causes relief due to the tingling sensation in the listener/watcher. In this study, new possibilities for nation branding will be discussed by making a theoretical reading on the communication power of ASMR in the cultural diplomacy dimension.
Using Plutchik’s wheel of emotions framework, we identify the emotional content of 133,487 social media posts and the audience’s emotional engagement expressed in 2,824,162 comments on those posts. We measure nine emotions (anger, anticipation, anxiety, disgust, joy, fear, sadness, surprise, trust) and two sentiments (positive and negative) using two extraction resources (EmoLex, LIWC) for eight major news outlets across four social media platforms (Facebook, Instagram, Twitter, and YouTube) during eight months. We then apply two approaches (Logistic Regression, Long Short-Term Memory) to predict emotional audience reactions before and after publishing the posts. Findings show significant differences for positive emotions but not for negative in the comments among the platforms. F1-scores for predicting emotional audience engagement are more than 70% for some emotions for some news outlets. Implications are that news outlets have leverage in steering emotional engagement for posts on social media platforms. The findings have theoretical and practical implications for understanding the complex emotional and informational interplay among social media content, platforms, and audiences.
Full-text available
طی سال‌های اخیر، استفاده شرکت‌های صنعتی از بازاریابی محتوایی دیجیتالی جهت حفظ و تعمیق روابط با مشتریان به میزان فزاینده‌ای افزایش یافته‌است. با این حال براساس یافته‌های مطالعات پیشین، اغلب بنگاه‌های اقتصادی نسبت به همسوسازی استراتژی بازاریابی محتوایی با اهداف رقابت‌پذیری خود بی‌توجه می‌باشند. ضمن آن‌که محققان تاکنون چارچوب جامعی جهت اجرای بازاریابی محتوایی همسو با اهداف رقابت‌پذیری بنگاه ارائه ننموده‌اند. از اینرو، پژوهش حاضر با هدف توسعه دیدگاه نظری نسبت به پیاده‌سازی بازاریابی دیجیتالی صنعتی تاثیرگذار و ارائه مدل بازاریابی محتوایی همسو با نیازهای مشتریان صنعتی و اهداف رقابت‌پذیری بنگاه اقتصادی انجام پذیرفت. برای این منظور، ابتدا مرور مفهومی و جامع درخصوص ادبیات پژوهش انجام شد. در گام بعدی، عوامل تاثیرگذار در موفقیت بازاریابی محتوایی در بازار صنعتی استخراج گردید. سپس، ارکان کلیدی مدل بازاریابی محتوایی همسو با اهداف رقابت‌پذیری بنگاه اقتصادی شناسایی شد. پس از آن، مدل مفهومی پژوهش ارائه گردید. درنهایت، صحت یافته‌ها در مدل ارائه شده با استفاده از روش کمترین مربعات جزئی در نرم افزار اسمارت پی.ال.اس، مورد بررسی قرار گرفت. یافته‌های این پژوهش، مبین لزوم اتخاذ رویکرد تلفیقی نسبت به نیازهای بازار هدف و همچنین اهداف رقابت‌پذیری بنگاه اقتصادی جهت تبیین استراتژی بازاریابی محتوایی اثرگذار می‌باشد.
Famous video bloggers (vloggers) on YouTube can develop large audiences, which can be related to the gaining of audience engagement (AE), manifested by the viewers’ participation and consumption on YouTube. Studies have unveiled vloggers’ behaviors for engaging audiences, or audience engagement behaviors (AEBs), in their videos, including interacting with viewers via comments, disclosing self-information, giving rewards, and offering other information. Meanwhile, video blogs (vlogs) are produced under “vlogging context” - situational elements involved in vlog production. Studies have shown the effect of context on the content of online media. However, while it can be argued that context can affect vlog content produced, the contextual factors that may shape vloggers’ AEBs within the content have not been explicitly explored. This research aims to propose contextual factors that can condition vloggers’ AEBs on YouTube. A qualitative case study on three popular vloggers was implemented. A thematic analysis was performed on sampled vloggers’ videos to identify contextual factors that can condition the three vloggers’ AEBs. The results propose that personal, environmental, and medium context are three main contextual factors that condition the three vloggers’ AEBs. This research argues that how vloggers’ AEBs are presented to the audience depends on their vlogging context. It expands the understanding of YouTube vloggers or similar streaming media creators’ practices for AE by considering the role of context.
In a crowded social media marketplace, restaurants often try to stand out by showcasing elaborate “Instagrammable” foods. Using an image classification machine learning algorithm (Google Vision AI) on restaurants’ Instagram posts, this study analyzes how the visual characteristics of product offerings (i.e., their food) relate to social media engagement. Results demonstrate that food images that are more confidently evaluated by Google Vision AI (a proxy for food typicality) are positively associated with engagement (likes and comments). A follow-up experiment shows that exposure to typical-appearing foods elevates positive affect, suggesting they are easier to mentally process, which drives engagement. Therefore, contrary to conventional social media practices and food industry trends, the more typical a food appears, the more social media engagement it receives. Using Google Vision AI to identify what product offerings receive engagement presents an accessible method for marketers to understand their industry and inform their social media marketing strategies.
The digital era has introduced many changes in the consumer marketplace. Social media and especially social networking sites redefined how consumers relate to and behave towards brands, as well as the brand-consumer relationship. Within this context and the heightened resistance to brand communication through traditional media, marketeers are turning to other strategies to connect with their customers and influence their consumer journey. One of these strategies is influencer marketing. In the last years, brands have used social media influencers as endorsers of their products and services, and as brand ambassadors. Digital influencers connect consumers and brands, strengthening their bond and allowing the brand to reach their target in a more natural way to influence the consumer buying process. In this chapter we will provide a narrative review on the role of digital influencers on the consumer decision processes.
Full-text available
While current literature has sufficiently profiled word-of-mouth (WOM) marketing, customer relationship management, brand communities, search engine optimization, viral marketing, guerilla marketing, events-based marketing, and social media each on an isolated, individual basis, there is no comprehensive model that effectively incorporates all of these elements. The first purpose of this paper is to therefore profile the current literature landscape surrounding WOM marketing, alternative marketing communications, and social media as viable components of integrated marketing communications. Additionally, this paper aims to develop an integrated alternative marketing communication conceptual model that can be applied by industrial practitioners to help them achieve their marketing objectives.
Full-text available
Size is the first, and at times, the only dimension that leaps out at the mention of big data. This paper attempts to offer a broader definition of big data that captures its other unique and defining characteristics. The rapid evolution and adoption of big data by industry has leapfrogged the discourse to popular outlets, forcing the academic press to catch up. Academic journals in numerous disciplines, which will benefit from a relevant discussion of big data, have yet to cover the topic. This paper presents a consolidated description of big data by integrating definitions from practitioners and academics. The paper's primary focus is on the analytic methods used for big data. A particular distinguishing feature of this paper is its focus on analytics related to unstructured data, which constitute 95% of big data. This paper highlights the need to develop appropriate and efficient analytical methods to leverage massive volumes of heterogeneous data in unstructured text, audio, and video formats. This paper also reinforces the need to devise new tools for predictive analytics for structured big data. The statistical methods in practice were devised to infer from sample data. The heterogeneity, noise, and the massive size of structured big data calls for developing computationally efficient algorithms that may avoid big data pitfalls, such as spurious correlation.
Full-text available
Photos are becoming prominent means of communication online. Despite photos' pervasive presence in social media and online world, we know little about how people interact and engage with their content. Understanding how photo content might signify engagement, can impact both science and design, influencing production and distribution. One common type of photo content that is shared on social media, is the photos of people. From studies of offline behavior, we know that human faces are powerful channels of non-verbal communication. In this paper, we study this behavioral phenomena online. We ask how presence of a face, it's age and gender might impact social engagement on the photo. We use a corpus of 1 million Instagram images and organize our study around two social engagement feedback factors, likes and comments. Our results show that photos with faces are 38% more likely to receive likes and 32% more likely to receive comments, even after controlling for social network reach and activity. We find, however, that the number of faces, their age and gender do not have an effect. This work presents the first results on how photos with human faces relate to engagement on large scale image sharing communities. In addition to contributing to the research around online user behavior, our findings offer a new line of future work using visual analysis.
Full-text available
We investigate the effect of social media content on customer engagement using a large-scale field study on Facebook. We content-code more than 100,000 unique messages across 800 companies engaging with users on Facebook using a combination of Amazon Mechanical Turk and state-of-the-art Natural Language Processing algorithms. We use this large-scale database of advertising attributes to test the effect of ad content on subsequent user engagement − defined as Likes and comments − with the mes-sages. We develop methods to account for potential selection biases that arise from Facebook's filtering algorithm, EdgeRank, that assigns posts non-randomly to users. We find that inclusion of persuasive content − like emotional and philanthropic content − increases engagement with a message. We find that informative content − like mentions of prices, availability and product features − reduce engagement when included in messages in isolation, but increase engagement when provided in combination with persuasive attributes. Persuasive content thus seems to be the key to effective engagement. Our results inform advertising design in social media, and the methodology we develop to content-code large-scale textual data provides a framework for future studies on unstructured natural language data such as advertising content or product reviews.
Instagram is a relatively new form of communication where users can easily share their updates by taking photos and tweaking them using filters. It has seen rapid growth in the number of users as well as uploads since it was launched in October 2010. In spite of the fact that it is the most popular photo capturing and sharing application, it has attracted relatively less attention from the research community. In this paper, we present both qualitative and quantitative analysis on Instagram. We use computer vision techniques to examine the photo content. Based on that, we identify the different types of active users on Instagram using clustering. Our results reveal several insights about Instagram which were never studied before, that include: 1) Eight popular photos categories, 2) Five distinct types of Instagram users in terms of their posted photos, and 3) A user's audience (number of followers) is independent of his/her shared photos on Instagram. To our knowledge, this is the first in-depth study of content and users on Instagram. Copyright © 2014, Association for the Advancement of Artificial Intelligence ( All rights reserved.
A considerable amount of social science research suggests an individual's initial perception of and reaction to another individual are affected by the physical attractiveness of the other person. The authors attempt to assess whether this general finding applies to people's perceptions of advertisements. Specifically, they assess the impact of attractiveness of male and female models on subjects' evaluations of ads, and seek to determine whether the reactions depend on the sex of the ad reader or on the type of product being advertised.
Social media is achieving an increasing importance as a channel for gathering information about products and services. Brands are developing its presence in social networking sites to meet brand awareness, engagement and word of mouth. In this context, the analysis of the factors that are conditioning consumer interaction with branded content becomes a matter of interest. This paper aims to shed light on those factors that are expected to impact on Facebook branded post popularity. A conceptual model is developed to reflect the influence of the content's richness and time frame on the number of comments and likes. An empirical analysis using multiple linear regressions is conducted based on 164 Facebook posts gathered from the fan pages of 5 Spanish travel agencies. Results suggest that the richness of the content (inclusions of images and videos) raises the impact of the post in terms of likes. On the other hand, using images and a proper publication time are significantly influencing the number of comments, whereas the use of links may decrease this metric. This study empirically contributes to the existing literature on the management of marketing strategies for consumer engagement in social networking sites.
Conference Paper
Over the past decade, social network sites have become ubiquitous places for people to maintain relationships, as well as loci of intense research interest. Recently, a new site has exploded into prominence: Pinterest became the fastest social network to reach 10M users, growing 4000% in 2011 alone. While many Pinterest articles have appeared in the popular press, there has been little scholarly work so far. In this paper, we use a quantitative approach to study three research questions about the site. What drives activity on Pinterest? What role does gender play in the site's social connections? And finally, what distinguishes Pinterest from existing networks, in particular Twitter? In short, we find that being female means more repins, but fewer followers, and that four verbs set Pinterest apart from Twitter: use, look, want and need. This work serves as an early snapshot of Pinterest that later work can leverage.
Conference Paper
Large Convolutional Neural Network models have recently demonstrated impressive classification performance on the ImageNet benchmark \cite{Kriz12}. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we address both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. We also perform an ablation study to discover the performance contribution from different model layers. This enables us to find model architectures that outperform Krizhevsky \etal on the ImageNet classification benchmark. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.