Wu2013 Article SocialMediaResearchAReview

J Syst Sci Syst Eng (Sep 2013) 22(3): 257-282 ISSN: 1004-3756 (Paper) 1861-9576 (Online)
DOI: 10.1007/s11518-013-5225-6 CN11-2983/N
SOCIAL MEDIA RESEARCH: A REVIEW
Junjie WU1 Haoyan SUN2 Yong TAN2

1
School of Economics and Management, Beihang University, Beijing 100191, China
wujj@buaa.edu.cn
2
Michael G. Foster School of Business, University of Washington, Seattle, Washington 98195-3226, U.S.A.
haoyan@uw.edu, ytan@uw.edu ()
Abstract
Social media is fundamentally changing the way people communicate, consume and collaborate. It
provides companies a new platform to interact with their customers. In academia, there is a surge in
research efforts on understanding its effects. This paper aims to provide a review of current status of
social media research. We discuss the specific domains in which the impacts of social media have been
examined. A brief review of applicable research methodologies and approaches is also provided.
Keywords: social media, empirical models, experimental methods, analytical approaches, predictive
analytics
1. Introduction Different social media platforms play

There is a growing consensus that social different roles in sharing information,
media is fundamentally changing the way people connecting people, and creating knowledge.
communicate, consume and collaborate. Within However, roles have evolved over time to suit
just a few years, social media has integrated the needs of social interactions and business
itself into almost every aspect of our personal strategies. Facebook and Twitter, traditionally
and professional lives, and as a result, has designed to connect people, express opinions
brought a lot of changes on businesses in and re-kindle real life social relationships, have
promoting sales, gaining customers and now been widely used as a marketing tool to
establishing trusts. Managers try to identify gain brand awareness or obtain customer
ways in which firms can make profitable use of feedbacks. Indeed, Kumar et al. (2013) show
social media applications such as Blogs, that social media can be used to generate growth
YouTube, Facebook, and Twitter. in sales, return on investment, and positive word

This research was partially supported by the National Natural Science Foundation of China (NSFC) under Grants
71322104,71171007,70901002 and 71031001, by the National Information Security Research Plan of China under
Grant 2012A137, by the Foundation for the Author of National Excellent Doctoral Dissertation of PR China under
Grant 201189, and by the Program for New Century Excellent Talents in University under Grant NCET-11-0778.
Dr. Yong Tan was supported in part by NSFC under Grants 71328103 and 71231002.
 Systems Engineering Society of China and Springer-Verlag Berlin Heidelberg 2013

WU, SUN & TAN: Social Media Research: A Review
258 J Syst Sci Syst Eng
of mouth, and can spread brand knowledge way in which information is created and
further. Luo et al. (2013) find a significant exchanged on those applications, such as:
predictive relationship between social media and  E-commerce sites with user-generated
firm equity value. Social media-based metrics content feature (e.g., Amazon.com )
(Web blogs and consumer ratings) are observed  Content sharing sites (e.g., YouTube,
to be significant leading indicators of firm Flicker, blogs)
equity value, stronger than conventional online  Social networking sites (e.g., Facebook,
behavioral metrics (Google searches and Web Twitter, Linkedin)
traffic).  Virtual community (e.g., online gaming,
As technology and innovation are moving Second Life)
forward, there are a lot more social and  Collaborative projects (e.g., Wikipedia,
physiological patterns that can be observed from Dropbox, Google Docs)
social media. Besides helping organizations to Through examining major streams in social
guide, promote, and shape online conversations, media research, we hope to provide a broad
social media has also worked as a useful tool to research agenda on social media and social
consumers when making decisions or gaining networks that highlights the contributions made
knowledge. For example, consumers have till now and stimulates interest in future
proved to learn from reviews more effectively research endeavors.
than from their own experience with other books The rest of the paper is organized as follows.
of the same genre (Zhao et al. 2013). Customer In next section, we discuss the contexts under
influences, through social media platforms, which social media research have been
grow exponentially and extend beyond conducted. This is followed by discussions on
geographic locations and individual preferences. empirical, experimental, and analytical models
Therefore, it is prudent for firms to engage for descriptive analytics. Then, we review the
customers in their social media brand research methodologies in the area of predictive
communities. Such customer involvements have analytics. Finally, we offer some concluding
shown to lead to a positive increase in purchase remarks.
expenditure (Goh et al. 2013). Rishika et al.
(2013) find that customer participation in a 2. Genres of Social Media Research
firm’s social media efforts leads to an increase in Social media has given companies access to
the frequency of customer visits. unprecedented volumes of information about
The social media phenomena have also their clients and purchasing behaviors on an
attracted attention from researchers. For aggregate level. It has also integrated itself into
example, Information Systems Research journal uses that go way beyond the familiar
has recently published a special issue on social applications for sales, marketing and community
media (Aral et al. 2013). In this paper, we aim to building. The way social media is rolled out has
provide a review of recent studies related to been evolving towards different goals; internal
social media applications, based on the primary networks, mobile applications, social-care,etc.
J Syst Sci Syst Eng 259
have all started embracing social media tools. Review information on certain products is
The challenge, which confronts everyone from widely available from online stores like
small businesses to researchers on social media, Amazon.com or any sites carrying user
is how to leverage all of the information and generated content. People have found different
turn them into actionable policies. In this section, stories regarding the factors that influence the
we introduce some popular research topics on sales. Chevalier and Mayzlin (2006) find that
social media. reviews with positive valence have positive
Word-of-Mouth. The availability of data impact on sales. By contrast, Liu (2006) states
from fast changing e-commerce environment that this explanatory power of sales comes from
has created exciting opportunities for the volume of WOM but not from its valence.
researchers to answer questions that have direct Sonnier et al. (2011) demonstrate a significant
and critical impact on businesses and consumers. effect of positive, negative, and neutral online
A number of researches have studied the communications on daily sales performance, and
network impact brought by social media. One find that aggregating online comments over
question that has been addressed by many valence masks the effect on sales. Later on,
scholars in various contexts is whether the many studies started to focus on the nature of
word-of-mouth effect has impact on sales and products. Sun (2012) studies the impact of rating
how the effect brings the changes to the sales variance on product sales and finds that a higher
figure. Traditionally, the impact of variance is usually associated with a niche
word-of-mouth was limited to the people close product and will lead to a higher demand if and
to each other and interacting with on a daily only if the average rating is low. Gu et al. (2012)
basis; but nowadays, word-of-mouth has discover that external WOM has a significant
endowed more power with social media and has impact on the sales of high-involvement
shaken off the boundary completely. Therefore, products that consumers spend considerable
the accessibility of online word-of-mouth time searching for before purchasing.
information has made it possible to study online Even though a great amount of previous
conversations in all aspects. The numerous studies have been discussing the impact of social
experimental and empirical studies in the media or what is shaped by social network
information systems and marketing literature effects, there are also more and more researches
have shown the importance of social media in talking about the processes or factors that drive
this respect. those consequences, especially the drivers of the
The most direct indicator of word-of-mouth word-of-Mouth effects. Godes and Mayzlin
– online ratings and reviews – has been well (2004) show that a measure of the dispersion of
acknowledged to have impacts on sales, while online conversations across communities has
the dynamics of impacts brought by the explanatory power in a dynamic model of TV
ratings/reviews are still debatable. Many of the ratings. Moreover, Godes and Mayzlin (2009)
researchers studying the impact of online initiate a field study to evaluate the effectiveness
conversations are limited to the data they obtain. of the commonly used opinion leader
designation and find that although opinion movie performances.

leadership is useful in identifying potentially A recent study by Lee et al. (2013) examines
effective spreaders of WOM among very loyal how users generate online reviews in the
customers, it is less useful for the sample of less presence of social influence. Using a data set on
loyal customers. Berger and Schwartz (2011) movie box office and user online social
examine how product characteristics shape networking behaviors, they find the existence of
immediate and ongoing WOM from a herding behaviors for popular movies, that is, a
perspective of product characteristics and user is more likely to give a higher rating if the
physiological factors, and suggest that rather prior aggregated user rating is higher. In doing
than being driven by motivation, WOM, so, they have carefully removed the confounding
particularly ongoing WOM, is driven by effect stemming from homophily (that users may
accessibility. People tend to talk about whatever have similar preferences and tastes). Ho et al.
comes to their minds, regardless of how (2013) discover that a user is more likely to
mundane it may be. Moe and Schweidel (2012) provide a rating for a product or service if the
find significant heterogeneity with respect to gap between her expected and realized
consumers’ desire to post in high-consensus perceptions of quality is larger, and thus
versus high-variance environments (Godes and explained the commonly observed U-shape
Silva 2012). distribution of online ratings.
Although word-of-mouth is the most direct Information Diffusion and Network
way for us to see the impact of social media on Dynamics. Emotion can often spread rapidly
sales, there are other innovative ways people through large crowds, and thus, individual ideas
have developed based on the digital structure of or behaviors are very much influenced by
online networks in order to push for more sales. someone else. Social networking sites like
Jing and Xie (2011) argue that a “Group Twitter, Facebook, Pinterest, and etc. have been
Buying” strategy, under which consumers enjoy used more and more often to broadcast people’s
a discounted group price if they are willing and opinion and ideas with the world, through ways
able to achieve a required group size and such as “like”, “re-tweeting”, or “re-pinning”.
coordinate their transaction time, actually boosts Understanding and predicting the outcomes of
sales for sellers. Oestreicher-Singer and those complex interactions is, therefore, a topic
Sundararajan (2012) find that co-purchase of great interest.
relationship leads to higher demand. However, In the business world, people care about
the tremendously increasing comments and social contagion because they want to
reviews over time also suggest that too much understand how their products can go
information may lead to customer information viral. Diffusion models can be useful for
overload and thus reviews may exhibit a managers to forecast demand and to plan a
declining pattern over time (Godes and Silva variety of pre-launch and post-launch strategic
2012). Gopinath et al. (2013) examine how local decisions such as the optimal level of product
geographic markets affect the impact of blogs on sampling, optimal pricing, and optimal timing of
successive generations of a product (Hosanagar formed, this similarity evolved in different ways
et al. 2010) There is an emerging stream of in different subgroups of dyads. Fang et al.
literature in which the impact of diffusion on (2013) take a data-driven approach to examine
individuals’ adoption is assessed. For instance, user adoption in a social network. They identify
Iyengar et al. (2011) find that the amount of key factors that underlie a social entity’s
contagion is moderated by both the recipients’ adoption decision and operationalize them using
perception of their opinion leadership and the Naïve Bayesian learning for accurate prediction.
sources’ volume of product usage. Because of Internal Social Network. There are more
the importance of opinion leaders, identifying and more organizations using social media tools
opinion leaders in a community is an essential as part of their internal communications, with
job for the purpose of marketing. Trusov et al. the aim to help engage employees on key topics
(2010) develop an approach to determine the such as performance, collaboration, culture and
opinion leaders in a social network by using the values. As the popularity of social media grows
longitudinal records of members’ log-in activity. tremendously, companies are getting more and
Besides general information diffusion, various more interests in understanding the value of
questions have been asked such as the factors in internal social network. IBM, for instance, has
determining peer influence and contagion. Aral been doing research to see how social
and Walker (2011) design a field experiment to networking can impact enterprise networks, and
examine how firms can create word-of-mouth found that many companies are encouraging
peer influence and social contagion by designing employees to use their social networks so they
viral features into their products and marketing can connect with other employees, help people
campaigns. They find that active-personalized socialize when they take a break, or even help
viral messages are more effective in encouraging contribute to other work-related issues (DiMicco
adoption per message, while passive-broadcast et al. 2008). A few studies have been conducted
messaging is used more often, generating more in understanding the true impact of internal
total peer adoption in the network. social network on employee behaviors and
On the other hand, it is also important to internal network structure. For example, Wu
understand the dynamics of digital content (2013) studies the change in employees’ network
diffusion structured through a network. Susarla positions before and after the introduction of a
et al. (2012) study the networked structure of social networking tool, and finds that
social influence through a Youtube data set and information-rich networks enabled through the
find the evidence that interactions in networked use of social media can drive both work
structure can guide opinion formation and direct performance and job security, although there is a
product discovery. Zeng and Wei (2013) observe trade-off between engaging in social
from a Flicker dataset that around the time of the communication and gathering diverse
formation of a social tie, members of dyads information. Sasidharan et al. (2012) examine
began to upload more similar photos than they the roles that the social network structures play
did before that time. After a social tie was in influencing the ERP post-implementation
success. They find that centralized structures compared to computer-based internet users.
inhibit implementation success, while Ghose and Han (2011) analyze whether there is
employees with high in-degree and betweenness a positive or negative interdependence between
centrality revealed high task impact and the mobile-phone-based content generation
information quality. behavior and content usage behavior.
Mobile. Social networking once began in the Healthcare. Information technology has yet
online space, but it has rapidly spread to mobile to revolutionize healthcare the way it has
platforms. More people are using smartphones changed our daily lives. Nowadays a cadre of
and tablets to access social media so that social platforms, such as PainentsLikeMe and
consumption of mobile data usage is being Patient Fusion, aim to disrupt the way people
driven by mobile social networking. According share information about personal health,
to Nielson’s 2012 social media report, 46% of physicians, and treatments. The intersection of
social media users use their smartphones to healthcare and social media represents a
access social media, and the time spent on promising space for future information systems
mobile apps and mobile web account for 63% of research (Fichman et al. 2011). Although social
the year-over-year growth in overall time spent media community has grown tremendously in
on social media. Besides native mobile social every industry, there are still very few researches
networks, such as Foursquare, Instagram, and that systematically study the social influence or
Path that are built around mobile functionality, peer-to-peer sharing among patients in online
almost all the social networking websites have healthcare communities. As social media is
created mobile apps to give their users instant utilized more and more by hospitals and medical
and real-time access from their devices. Mobile professionals as a means to convey general
social media makes use of its GPS location health information, it is of interest to see the
feature and time flexibility to influence an network impact or diffusion model to be applied
upward trend in the popularity and accessibility in the healthcare community. There exist a few
of e-commerce, or offline spending. So far, the studies conducted from different perspectives to
line between mobile and web is becoming very confirm the social impact on healthcare. For
blurred. Given the current trend of instant example, Miller and Tucker (2013) collect data
communication, location-based services, and on whether the hospital is actively managing its
augmented reality, everything requires mobile social media presence by customizing their
devices and technology. It is foreseeable that Facebook page. They find that when an
mobile and web-based social networking organization actively manages its social media
systems will work symbiotically to spread presence, it predominantly succeeds in
content, increase accessibility and connect users increasing user-generated content as a function
from wherever they are. In their recent research, of number of their employees, rather than the
Ghose et al. (2012) find that ranking effects and number of clients. Angst et al. (2010) use a
the benefit of browsing for geographically close social contagion lens to study the dynamic,
matches are both higher on mobile phones temporal process of the diffusion of electronic
medical records in the population of U.S. an idea the organization finds valuable enough
hospitals and find that diffusion can be to implement but are unlikely to repeat their
accelerated if specific attention is given to early success once some of their ideas are
increasing social contagion effects in the case of implemented.
hospitals’ likelihood of adopting electronic Kleemann et al. (2008) characterize different
medical records when subjected to the influence types of crowdsourcing given different levels of
of prior adopters. Yan and Tan (2013) analyze an knowledge companies expect the consumers to
online healthcare community and find the contribute. For example, companies may ask for
evidence that social support exchanged among comments and suggestions on product
patients has a positive impact on improving their development and configuration, or advices to
health conditions. Yan et al. (2013) study how develop a whole new product, or any new
patients identify and select proper users to information or documentation, and etc. Among
communicate in the healthcare community to those levels, the most common one is what
reduce their information overload and to receive nowadays adopted by almost all online vendors
expected social gains from the online activities. – product ratings and reviews. For instance,
Crowdsourcing. In general, all different Archak et al. (2011) capture textual data of
kinds of user-generated content can be Amazon reviews, and show how it can be used
considered as crowdsourced content as these for learning consumers’ relative preferences for
content provides useful information and different product features and for predictive
wisdoms to some extent. A few researches have modeling of future changes in sales.
been conducted to access the value of mining A recent study by Hildebrand et al. (2013)
consumer reviews and ratings. As user content points out that social media could be bad for
generation platforms have been widely crowdsourcing. The negative influence of
developed, many organizations are now using consumer feedback on their satisfaction could
the platform to outsource their ideation efforts in result from increased decision uncertainty and
facilitating their innovation process. This perceived process complexity.
“crowdsourcing” approach has been receiving Crowdfunding. Crowdfunding can be
lots of attention ever since, and several viewed as an element of crowdsourcing. In
organizations have implemented online particular, it deals with financial contribution to
crowdsourcing systems that gather ideas for new an entity, instead of contributing ideas to a
products and services from the “crowd” of company. There is an emerging stream of
nonexperts. One prominent example of the research that has studied the new
crowdsourcing sites is Dell’s IdeaStrorm. Bayus micro-financing method of peer-to-peer lending,
(2013) collects two years’ data from Dell’s or crowd-funding. Crowd-funding has been
IdeaStorm community to study the nature of a defined as “the financing of a project of a
crowdsourced idea generation process over time. venture by a group of individuals instead of
They find that serial ideators are more likely professional parties” (Schwienbacher and
than consumers with only one idea to generate Larralde 2010). Basically, people who want to
borrow are matched with those that want to lend linkage between marketing effort and the
on the web-based microlending platform. This success of crowd-funded projects.
out-of-the-box, non-traditional investing option
has become a very compelling alternative to 3. Descriptive Research
traditional investing, and quite a few Descriptive Research aims to uncover the
crowd-funding platforms have arisen over the underlying mechanisms in the problem under
last few years. The earliest successful example is study. The emphasis of descriptive research is on
Sellaband.com, a Dutch-based marketplace in identifying and accurately describing causal
which musical artists raise funds to produce and relationships. In general, we can categorize the
sell albums (Burtch et al. 2013). Prosper.com, methodologies into three areas: econometric or
founded in 2006, has grown quickly and become empirical, experimental, and analytical modeling.
a well-known crowd-funding site in the United In the following, we describe each area and give
States. As the big competitor of Prosper, some examples related to social media from
Lending Club has teamed up with Google and existing literature.
has grown really fast to compete with
institutional lenders. A few studies have done 3.1 Empirical Methods
based on the data obtained from these prominent Given that various forms of data exist in the
sites. Lin et al. (2013) examine Prosper.com in context of social media, econometric methods
an attempt to identify the network effects in are often the natural choice employed to derive
assisting individual’s crowd-funding decision insights. For empirical research, there are
making. They find that a borrower with higher typically two major issues to consider:
social capital is more likely to gain trusts from specification and identification. Model
the lender and thus has higher probability of specification concerns the selection of the
raising fund successfully. However, Zhang and variables included in the model. Very often, we
Liu (2012) find, counter to intuition, that need to provide justifications, from the related
herding effect happens among lenders when the economic and social theories, why certain
borrower exhibits signals of low quality. The variables or measures are included or excluded.
possible explanation to this is that lenders Non-experimental researches, whether using
assume there is some private information about reduced forms or structural models, are all
the borrower that they are not privy to. Burtch subject to an identification problem. A model
(2011) also provides the evidence in his research, has to be properly specified so that different
stating that herding behavior is a network characteristics are identifiable and
externality in a crowd-funding market. In distinguishable given the data. That is, the model
additional, Burtch et al. (2013) empirically study parameters for all the variables (or covariates)
social influence in a crowd-funded marketplace can be properly estimated.
for online journalism projects. They examine It is often the case that the data is imperfect.
both the antecedents and consequences of the The deficiencies in the data cause challenges in
contribution process and find that there is a clear model identification. One example is the omitted
variable problem resulting from the fact that product sales. The difference-in-difference
certain information is unobservable. If not strategy has also been used widely in the
properly handled, omitted variables can literature to control for common unobservable
introduce biased estimates of key model heterogeneity. Chevalier and Mayzlin (2006)
parameters, and hence lead to wrong, or at least adopt the difference-in- difference strategy to
inaccurate implications. Another consequence is eliminate a possible website-specific fixed effect,
the problem of endogeneity, that is, the error when studying the effect of WOM on book sales
term is correlated with some independent across different online bookstores. Similarly Sun
variables. This will happen when an omitted (2012) employs a difference-in-difference
variable is, for example, a decision variable strategy to a data from both Amazon and Barnes
which drives some independent variables. For & Noble to control for unobserved book
cross sectional data, most identification characteristics that may influence both sales and
strategies rest on identifying plausible ratings.
instruments. When measuring the impact of Social Effects. There are additional
on-line movie reviews with aggregated market challenges in dealing with network data to
level data, Chintagunta et al. (2010) use the identify endogenous social effects. This is
exogenous variables from previous markets as commonly known as the reflection problem, first
instruments for user ratings in subsequent pointed out by Manski (1993). This problem
market, in order to account for the sequential arises when a researcher is observing the
release of movies across geographic markets. distribution of behavior and tries to infer where
Their approach can be generalized in dealing the average behavior in some group influences
with endogeneity concerns to other settings the behavior of the individuals that comprise the
where products are sequentially rolled out. group. A linear specification where the
The availability of panel data provides a individual behavior is linearly related to the
better solution to the endogeneity problem. With group average will result in a tautology and the
panel data one can control product heterogeneity social influence is not identifiable. To resolve
through fixed effect approach, or control this problem, there are many approaches, for
unobserved error via random effect approach. example, using a nonlinear relationship which of
Both of the two approaches serve the role of course needs to be justified by underlying
picking up common components of a certain theories. Another approach is to explicitly
variable. Gu et al. (2012) collect a panel data to incorporate social structures and measures in the
investigate the role of internal and external model. Instead of using group average,
WOM sources on retailer sales of individuals in social relationships are modeled to
high-involvement products. Besides adopting a contribute differently.
fixed-effect model to control product This actually leads to a natural integration of
heterogeneity and brand effect, they also use an econometric and social networks analyses.
instrumental variable approach to control the Social networks analysis (Wasserman and Faust
endogeneity between the WOM volume and 1994) provides a comprehensive set of measures,
for example, various centralities, structural variables are unobservable, they describe needed
equivalence, structural holes, cohesive individual heterogeneity and also play very
subgroups, and etc. at both individual and important roles in driving some social behaviors.
network levels. As a social network is mapped Hence, it is especially useful to model them and,
with respect to a specific social relationship such if possible, make an inference to uncover them.
as friendship, subscription, co-worker, and so on, The latent space model, developed by Hoff et al.
the estimated effects can be better interpreted in (2002), is a perfect example. In this model, each
the proper context. actor is assumed to have a position in a social
In general, it is agreed that it is difficult, if space. The positions are unknown, so the social
not impossible, to cleanly identify social space is a latent one. However, the probability of
influences. There could exist many confounding a link between two actors depends on the
effects, most notably, homophily and exogenous distance between them in the social space. If we
shocks. For example, if we observe two users can observe how actors communicate, form or
who downloaded the same mobile app, we can sever links, we can estimate their hidden social
say that maybe the second user was influenced positions.
socially by the first user, and hence followed. Similarly, the Hidden Markov Model (HMM)
However, it could also be the case that both (Singh et al. 2011) is a useful model to
users had similar preferences or tastes, or they understand latent drivers of social or economic
are homophilous. The third reason could be that behaviors. It consists of two Markov chains; one
they were exposed to the same exogenous shock, is observable while another is not. But these two
such as a marketing promotion. In order to tease chains are interrelated – the behavior observed
out each cause, one has to exert effort to identify in one chain is driven or influenced by the latent
and make use of their unique characteristics one. Once the model parameters are estimated,
which may be contextual. Bramoullé et al. (2009) either by maximum likelihood or Bayesian,
suggest a way to identify the true effect of social posterior analysis can be conducted to uncover
influence by accounting for the homophily. A the hidden chain. POMDP (partially observed
network specific unobservable is introduced to Markov decision process) is a variant of HMM
the individuals that belong to the same social which allows some parts of the otherwise
network structure. Then, we use group means of completely hidden chain to be observed. This
the variables and subtract the variables from model has also found applications in social
corresponding group means, to remove the effect media research (Yan and Tan 2013).
stemming from similar preferences. Recently, there is a new development in
Latent Variable Models. There are numerous applying latent instrument variable to resolve
econometric models that can be applied to potential endogeneity biases (Ebbes et al. 2009).
analyze social media data. A few of them are This method takes the advantage of the
particularly suited, for example, a class of distribution of data, and separates an
generalized latent variable models endogenous variable into exogenous and
(Rabe-Hesketh and Skrondal 2004). While some endogenous parts respectively.
Dynamic Models. Snijders (2006) has conduct policy analyses and produce
proposed a dynamic model of a dichotomous counterfactuals to answer some important
social network based on Markov Chain Monte “what-if” questions.
Carlo (MCMC), to study the formation and
evolution of a social network. The continuous 3.2 Experimental Methods
time Markov transition probabilities are Field experiments have been used in social
modeled as a function of a set of covariates media research to exogenously vary the
derived from individual and dyadic treatment across subjects. This helps to solve the
characteristics, for example, degree of ties, potential endogeneity problem. Another
reciprocity, transitivity, and etc. With the common problem in empirical research is that
observation of some snapshots of the network at sometimes what the data and/or model show is
some time points, the dynamic model can merely correlation rather than causality. It is
estimate the parameters associated with the therefore useful to carefully design experiments
covariates or factors which drive the network to identify causal relationships.
evolution. Aral and Walker (2011) design a field
Shriver et al. (2013) conduct an empirical experiment to examine how firms can create
study to confirm that users’ online content word-of-mouth peer influence and social
generation activity is codetermined by their contagion by designing viral features into their
social ties. In another study by Zeng and Wei products and marketing campaigns. Toubia and
(2013), the interaction between social ties and Stephen (2013) analyze and contrast two types
user content generation is also examined. They of user utility: intrinsic and image-related, in
find that members of dyads were more likely to motivating them to post content. They conducted
upload similar photos when they formed a tie; a field experiment to address an issue that the
however, the dyads of users with similar number of followers on Twitter could be
popularity began to upload less similar photos. endogenous by exogenously adding followers to
Structural Modeling. We have observed an a set of users. Zhang and Zhu (2011) take
increase in applying structural models in advantage of a natural experiment to examine
marketing and information systems research the causal relationship between group size and
(Chintagunta et al. 2006). Structural models rely incentives to contribute in the setting of Chinese
on economic or related theories to derive model Wikipedia, the Chinese language version of an
specification. This approach is especially suited online encyclopedia that relies entirely on
to capture the strategic behaviors of rational voluntary contributions. Claussen et al. (2013)
users or firms that maximize their respective make use of a rule change by Facebook to study
objects such as utility or profit. A dynamic the effects of rewarding user engagement.
model allows describing behaviors such as
3.3 Analytical Methods
forward-looking decision making. Structural
The analytical models in social media
models can resolve the potential endogeneity
research typically originate from microeconomic
problem. Another advantage is its capability to
theory and/or operations research. Recently, variety of social networks, EGP is polynomially
there are applications of the methodologies in solvable whereas PP is strongly NP-hard. They
other disciplines (for example, physics) in social find the solution of PP on several special
media, especially social networks, research. In networks: bi-cliques, balanced and full d-trees,
the following, we discuss a few papers with paths, cycles, and cliques, and proposed a
different modeling approaches. heuristic for general trees.
Game Theoretical Model. Mayzlin and Ising Spin Glass Model. The Ising model
Yoganarasimhan (2012) formulate game originated in statistical physics as a means of
theoretical model to explain why a rational modeling the interactions among electron spins
blogger may choose to link to another blog. It is to explain the emergence of magnetization. It is
assumed that bloggers differ by their abilities to often used to demonstrate phase transition and
break news and to find news in other blogs, critical phenomena. The Ising model has also
respectively. By linking, a blog signals to the been used to explore the emergence of critical
reader that it will be able to direct her to news in dynamics such as bubbles and crashes in various
other blogs in the future. However, the tradeoff economic settings. Recently, this model has been
is that linking generates a positive signal on the applied to examine herding in open source
rival’s news-breaking ability. They show that software development community (Oh and Jeon
linking will be in equilibrium when the 2006). Less attention has been paid, however, to
heterogeneity on the ability to break news is low how Ising-like structures can be used to model
relative to the heterogeneity on the ability to find the dynamics of opinion formation and product
news in other blogs. Overall, it is found that the adoption. The simplicity of the Ising
activity of linking enhances readers’ learning. methodology can help articulate models that are
Dou et al. (2012) investigate how software firms analytically tractable and empirically verifiable.
should optimize the strength of network effects There exists a simple mapping in that an
by adjusting the level of embedded social media electron can be considered as an individual in a
features. This paper analyzes the right market social network, and the spin which could be
seeding and pricing strategies in the presence of positive or negative represents the sentiment of
seeding disutility, with a focus on the interaction this individual’s opinion. The temperature in
between building more social media features to Ising model can be interpreted as the strength of
increase the strength of network effects and social interaction, whereas the external magnetic
seeding the market. field can be attributed to exogenous marketing
Operations Research Model. Dawande et al. effort. As such, some results in Ising model can
(2012) formulate optimization problems to be readily applied to social media. For example,
search for useful structures in social networks. the characteristics of the critical point where the
They consider two problems: EGP (the elite consensus of opinions is formed can be obtained
group problem) and PP (the portal problem) in relation with social interactions and
derived, respectively, from the notions of exogenous shocks.
influence and centrality. It is shown that for a
4. Predictive Analytics social media mining research in some subfields

The data generated by various online social most related to descriptive analytics.
media is of massive amounts, unstructured, and Sentiment Analysis. As massive opinion data
dynamic in nature, which poses great challenges emerges, there is a pressing demand for
to the so-called big data analytics (Chen et al. sentiment analysis that could aggregate the
2012). As an inter-discipline springing up in overall public opinions and characterize the
recent years, data mining is often regarded as sentiment variations over a given period of time.
one of the best solutions to cope with these This information allows for perceiving the brand
challenges. It tries to discover useful or reputation or the-word-of-mouth for some
actionable rules from big data (Tan et al. 2005), products, and predicting the box office revenues
by establishing heterogeneous (may be (Asur and Huberman 2010) or stock market
distributed) databases to store and manipulate (Bollen et al. 2011).
data of various kinds using both SQL and The lexicon-based method is one solution to
noSQL techniques (Stonebraker 2010), and sentiment analysis. With a comprehensive and
designing predictive algorithms suitable for authoritative emotional lexicon, this method is
parallel or distributed computing on platforms easy to apply and works well in some domains,
such as Hadoop (White 2012). like finance (Das and Chen 2007). Moreover,
As a methodology for predictive analytics, many experimental results show that the
data mining can also boost the descriptive machine learning approach could also achieve
analytics of social media. On one hand, data very high accuracy (Pang et al. 2002), which
mining enables the objective characterization of promoted the combination of the two methods
social behaviors of an individual, a group, or an (Melville et al. 2009).
organization upon the rich online interactive In the context of social media, more
data. This is particularly valuable for studying information such as emoticons, hashtags and
how, for example, sentiments or authorities of network structures is available, which provides
individuals, can influence the performances of new thoughts for sentiment analysis. For
online businesses or finance. On the other hand, example, Read (2005) shows that using training
the deployment of various data mining data labeled by emoticons can greatly reduce the
techniques on social media sites, such as dependency of domain, topic and time in the
recommender systems and community detection, machine learning techniques. Zhao et al. (2012)
provides precious opportunities to probe propose an emoticon-based method for
exciting new research topics. For instance, multi-sentiment analysis of Chinese tweets,
researchers are getting more and more interested which shows potential for abnormal event
in understanding the major factors that lead to detection. Wang et al. (2011) use graph models
the real success of social recommendation, or based on hashtags co-occurrence relationship to
the underlying forces that drive the dynamic perform sentiment analysis. Guerra et al. (2011)
evolution of specific communities. In what use the endorsement network to measure the
follows, we briefly introduce the advances of bias of social media users toward a topic, and
adopt a transfer learning approach to compute propose different heuristics to speed up the
the polarity of posted contents. computation including shortest-path (Kimura
Social Influence Computing. Social and Saito 2006), cost-effective lazy forward
influence computing aims to find influential (Leskovec et al. 2007), maximum influence path
nodes such as authorities, opinion leaders, (Chen et al. 2010), LDAG (Chen et al. 2010),
gatekeepers or mavens, which could activate and SPIN (Narayanam and Narahari 2010).
other nodes in social networks as much as Despite many algorithms proposed for
possible. Motivated by applications in viral influence modeling, few of them can work for
marketing (Domingos and Richardson 2001), very large social networks. Several studies that
personalized recommendation (Song et al. 2007), extract influence cascade model parameters
feed ranking (Ienco et al. 2010), and the analysis from real datasets to generate influence graphs
of online social network (Weng et al. 2010), the (Anagnostopoulos et al. 2008, Tang et al. 2009,
study of influence propagation has caught Saito et al. 2010, Goyal et al. 2010) can be
tremendous attention in the last decades. regarded as the initial works along this line.
In the early research, Krackhardt (1992) and Zhao et al. (2013) point out that the whole
Granovetter (1973) reveal the impact of the network topology can usually be unavailable and
strength of strong and weak ties to social works, hence introduce an approach based on graph
respectively. Later, some topological properties sampling.
and network metrics like the number of Recommender Systems. Recommender
followers, PageRank value and the number of systems suggest a few items from many possible
retweets were used to rank and find the most choices to users by learning their profiles. In the
influential users (Kwak et al. 2010). Based on last two decades, many different types of
these, Bakshy et al. (2011) propose diffusion recommender systems have been developed like
tree to quantify the user influences. Note that the the collaborative-filtering based (Schafer et al.
rankings by different influence measures are 2007), content based (Debnath et al. 2008), and
often inconsistent, and PageRank usually fails to hybrid (Symeonidis et al. 2008) systems, most
produce high quality target sets dispersing of which exploit user-item rating matrix to
across the whole network (Pandit et al. 2012). generate recommendations.
The methods above only take the network Recently the development of recommender
topology into account. Recently, Domingos and systems has been advanced by the rapid growth
Richardson (2001) introduce propagation factors of social media. In addition to user ratings,
to identify influential users in a marketing area social media further provides social signals and
as a learning problem. Subsequently, Kempe et rich contexts like friend networks, social tags,
al. (2003) formulate the problem as a discrete trust, and locations. Recommender systems
optimization problem and prove that it is research has advanced into how to use
NP-hard if the influence propagation is based on heterogeneous information to improve
the independent cascade model or the linear recommendation accuracy. For instance, Jamali
threshold model. Several follow-up studies and Ester (2010) incorporate the mechanism of
trust propagation into the matrix factorization shortage of manually labeled data in shilling
model, and led to a substantial increase in attack classification, Wu et al. (2012) provide a
recommendation accuracy, especially for semi-supervised method to detect shilling
cold-start users. The work by Konstas et al. attacks, and demonstrate its effectiveness on
(2009) takes into account both the social enhancing the product recommendations in
annotations and friendships established among Amazon.
users, items and tags, and adopts the generic Community Detection. People within an
framework of random walk with restarts to online social community usually hold similar
provide a more natural and efficient way to opinions, have common interests, and exert
represent social networks. Ye et al. (2010) more influences to each other. So discovering
exploit the social and geographical such social communities can enhance the
characteristics of users and locations to realize analysis of opinions, the understanding of
location-based recommendation services. influences among users, and the performance of
Besides the real item recommendation, link intelligent recommender systems. The
or key node recommendation on social networks community detection problem is initially
also attracts much attention. For example, formulated as finding a good K-way crisp
Saez-Trumper et al. (2012) combine temporal partition (Kernighan and Lin 1970, Slater 2008,
attributes of nodes and edges of network with a Newman 2006, Newman 2004, Flake et al. 2002,
PageRank based algorithm to find the Girvan and Newman 2002). Since networks
trendsetters for a given topic. Backstrom and keep evolving with new events and individuals,
Leskovec (2011) study the problem of inferring much attention is then paid to community
interactions among existing members that are evolution to reveal community structures at
likely to occur in the near future. The work by sequential timestamps (Yang et al. 2009, Lin et
Dong et al. (2012) also studies the problem of al. 2009, Tang and Liu 2012). Furthermore,
link recommendation across heterogeneous people notice that communities are nested in
networks where a ranking factor graph model is nature, and thus have great interests in finding
used. overlapping communities (Palla et al. 2005,
Online recommender systems are vulnerable Lancichinetti et al. 2009, Zhang et al. 2007,
to malicious users who inject biased ratings to Nepusz et al. 2007). We here briefly review
manipulate online product recommendations, i.e., basic methods in these three areas.
the “shilling attacks” or “profile injection The crisp community detection methods can
attacks” (Burke et al. 2005, Lam and Riedl 2004, be classified into two categories, in terms of
O’Mahony et al. 2004). The related studies whether or not a global optimization objective is
mainly focus on the three subareas: the shilling available. The methods with global optimization
attack generation models (Mobasher et al. 2007), objectives typically consider the global topology
the detection metrics (Burke et al. 2006, of a network, and aim to optimize the objective
Williams 2006), and the classification methods function defined over a network partition. The
(Mehta and Nejdl 2009). Considering the differences between these methods ultimately
come down to the global criteria and the to multiple social groups. Overlapping
algorithmic heuristics. For instance, the cut community detection thus gains increasing
criterion and its variations have been used by interests, which aims to discovery partitions that
Kernighan-Lin algorithm (Kernighan and Lin are not necessarily disjoint. The famous Clique
1970) and spectral methods (Slater 2008), and Percolation Method (CPM) (Palla et al. 2005) is
the modularity (Q function) proposed by based on the concept of k-clique, with CFinder
Newman has been used in a great deal of (http://www.cfinder.org/) being a successful
algorithms such as FastNewman (Newman 2006, implementation. Many other methods have been
Newman 2004). The methods without global presented including the link partitioning method
optimization objectives typically employ a (Ahn et al. 2010), the local expansion and
bottom-up strategy to find communities. The optimization methods (Lancichinetti et al. 2009,
maximum flow community algorithm (Flake et Shen et al. 2009), the fuzzy clustering based
al. 2002) and the GN algorithm (Girvan and methods (Zhang et al. 2007, Nepusz et al. 2007),
Newman 2002) are just two examples. In recent and etc.
years, with the emergence of super-large social
networks from Facebook, Twitter, and etc., 5. Conclusions
community extraction rather than partition This paper provides a concise, and hence
becomes a very hot topic in this area (Zhao et al. inherently incomplete, review of research on
2011, Wu et al. 2013). The efficiency and social media. We have discussed extant prior
evaluation issues are the two main concerns. studies which cover a wide spectrum of topics,
The studies on community evolution can be ranging from word of mouth to the applications
also divided into two types. One employs a of social media in healthcare and mobile
straightforward two-step method, where static platforms. A summary of data driven approaches
analysis is applied firstly to the snapshots of the in the form of data and text mining has also been
network, and then community evolutions are provided. We have also reviewed quantitative
introduced to interpret the change of research methods and discussed the challenges
communities over time. The other one attempts researchers may face when applying these
to unify the processes of community detection methodologies.
and evolution. Dynamic stochastic block model There are many opportunities for researchers
is widely used for this purpose, aided by some to advance our understanding of social media
optimization methods such as Gibbs sampling and its effects on businesses. For example,
(Yang et al. 2009) and Expectation- existing analytical studies on diffusion have
Maximization (Lin et al. 2009). Meanwhile, often used mean-field approximation which
community evolution on multi-mode networks, ignores the underlying network topology. Given
consisting of different types of nodes and the importance of network structure, it is
relations, has also attracted some attention (Tang imperative to develop a framework which
and Liu 2012). explicitly accounts for it. It is also interesting to
In real life, a person usually has connections identify and investigate novel business models
and social phenomena emerging in the online [7] Asur, S. & Huberman, B.A. (2010).
platforms such as microblog sites. Many of them Predicting the future with social media.
may become test beds, unavailable before, for IEEE/WIC/ACM International Conference
researchers to re-examine existing social and on Web Intelligence and Intelligent Agent
economic theories. Technology (WI-IAT), 1: 492-499
[8] Backstrom, L. & Leskovec, J. (2011).
References Supervised random walks: predicting and
[1] Ahn, Y.Y., Bagrow, J.P. & Lehmann, S. recommending links in social networks.
(2010). Link communities reveal multiscale Proceedings of the 4th ACM International
complexity in networks. Nature, 466(7307): Conference on Web Search and Data Mining,
761-764 2011: 635-644
[2] Anagnostopoulos, A., Kumar, R. & Mahdian, [9] Bakshy, E., Hofman, J.M., Mason, W.A. &
M. (2008). Influence and correlation in Watts, D.J. (2011). Everyone's an influencer:
social networks. Proceedings of the 14th quantifying influence on twitter. Proceedings
ACM SIGKDD International Conference on of the 4th ACM International Conference on
Knowledge Discovery and Data Mining, Web Search and Data Mining, 2011: 65-74
2008: 7-15 [10] Bayus, B.L. (2013). Crowdsourcing new
[3] Angst, C.M., Agarwal, R., Sambamurthy, V. product ideas over time: an analysis of the
& Kelley, K. (2010). Social contagion and dell ideastorm community. Management
information technology diffusion: the Science, 59(1): 226-244
adoption of electronic medical records in U.S. [11] Berger, J. & Schwartz, E.M. (2011). What
hospitals. Management Science, 56(8): drives immediate and ongoing word of
1219-1241 mouth? Journal of Marketing Research,
[4] Aral, S. & Walker, D. (2011). Creating 48(5): 869-880.
social contagion through viral product design: [12] Bollen, J., Mao, H. & Zeng, X. (2011).
a randomized trial of peer influence in Twitter mood predicts the stock market.
networks. Management Science, 57(9): Journal of Computational Science, 2(1): 1-8
1623-1639 [13] Bramoullé, Y., Djebbari, H. & Fortin, B.
[5] Aral, S., Dellarocas, C. & Godes, D. (2013). (2009). Identification of peer effects through
Social media and business transformation: a social networks. Journal of Econometrics.
framework for research. Information 150(1): 41-55
Systems Research, 24(1): 3-13 [14] Burke, R., Mobasher, B., Williams, C. &
[6] Archak, N., Ghose, A. & Ipeirotis, P.G. Bhaumik, R. (2006). Classification features
(2011). Deriving the pricing power of for attack detection in collaborative
product features by mining consumer recommendation systems. Proceedings of the
reviews. Management Science, 57(8): 12th International Conference on Knowledge
1485-1509 Discovery and Data Mining, 2006: 542-547
[15] Burke, R., Mobasher, B., Zabicki, R. & effect of word of mouth on sales: online
Bhaumik, R. (2005). Identifying attack book reviews. Journal of Marketing
models for secure recommendation. Beyond Research, 43(3): 345-354
Personalization: Workshop on Next [23] Chintagunta, P.K., Gopinath, S. &
Generation Recommender Systems Venkataraman, S. (2010). The effects of
[16] Burtch, G. (2011). Herding behavior as a online user reviews on movie box office
network externality. ICIS 2011 Proceedings, performance: accounting for sequential
2011: 28 rollout and aggregation across local markets.
[17] Burtch, G., Ghose, A. & Wattal, S. (2013). Marketing Science, 29(5): 944-957
An empirical examination of the antecedents [24] Chintagunta, P., Erdem, T., Rossi, P.E. &
and consequences of contribution patterns in Wedel, M. (2006). Structural modeling in
crowd-funded markets. Information Systems marketing: review and assessment.
Research, 24(3): 499-519 Marketing Science, 25(6) : 604-616
[18] Calais Guerra, P.H., Veloso, A. & Meira, [25] Claussen, J., Kretschmer, T. & Mayrhofer,
J.W. (2011). From bias to opinion: a P. (2013). The effects of rewarding user
transfer-learning approach to real-time engagement: the case of facebook apps.
sentiment analysis. Proceedings of the 17th Information Systems Research, 24(1):
ACM SIGKDD International Conference on 186-200
Knowledge Discovery and Data Mining, [26] Das, S.R. & Chen M.Y. (2007). Yahoo! for
2011:150-158 Amazon: sentiment extraction from small
[19] Chen, H., Chiang, R.H.L. & Storey, V.C. talk on the web. Management Science, 53:
(2012). Business intelligence and analytics: 1375-1388
from big data to big impact. MIS Quarterly, [27] Dawande, M., Mookerjee, V.,
36(4): 1165-1188 Sriskandarajah, C. & Zhu, Y. (2012).
[20] Chen, W., Wang, C. & Wang, Y. (2010). Structural search and optimization in social
Scalable influence maximization for networks. INFORMS Journal on Computing,
prevalent viral marketing in large-scale 24(4): 611-623
social networks. Proceedings of the 16th [28] Debnath, S., Ganguly, N. & Mitra, P.
ACM SIGKDD International Conference on (2008). Feature weighting in content based
Knowledge Discovery and Data Mining, recommendation system using social
2010: 1029-1038 network analysis. Proceedings of the 17th
[21] Chen, W., Yuan, Y. & Zhang, L. (2010). International Conference on World Wide
Scalable influence maximization in social Web, 2008: 1041-1042
networks under the linear threshold model. [29] DiMicco, J., Millen, D.R., Geyer, W.,
Proceedings of the 10th IEEE International Dugan, C., Brownholtz, B. & Muller, M.
Conference on Data Mining, 2010: 88-97 (2008). Motivations for social networking at
[22] Chevalier, J.A. & Mayzlin, D. (2006). The work. Proceedings of the 2008 ACM
Conference On Computer Supported (2012). How is the mobile internet different?

Cooperative Work, 2008: 711-720 search costs and local activities. Information
[30] Domingos, P. & Richardson, M. (2001). Systems Research, forthcoming
Mining the network value of customers. [38] Ghose, A. & Han, S.P. (2011). An
Proceedings of the 7th ACM SIGKDD empirical analysis of user content generation
International Conference on Knowledge and usage behavior on the mobile internet.
Discovery and Data Mining, 2001: 57-66 Management Science, 57(9): 1671-1691
[31] Dong, Y., Tang, J., Wu, S. & Tian, J. [39] Girvan, M. & Newman, M.E. (2002).
(2012). Link prediction and recommendation Community structure in social and biological
across heterogeneous social networks. networks. Proceedings of the National
Proceedings of the 12th IEEE International Academy of Sciences, 99(12): 7821-7826
Conference on Data Mining, 2012: 181-190 [40] Godes, D. & Mayzlin, D. (2004). Using
[32] Dou, Y., Niculescu, M.F. & Wu., D.J. online conversations to study word-of-mouth
(2013). Engineering optimal network effects communication. Marketing Science, 23(4):
via social media features and seeding in 545-560
markets for digital goods and services. [41] Godes, D. & Mayzlin, D. (2009).
Information Systems Research, 24(1): Firm-created word-of-mouth communication:
164-185 evidence from a field test. Marketing Science,
[33] Ebbes, P., Wedel, M. & Böckenholt, U. 28(4): 721-739
(2009). Frugal IV alternatives to identify the [42] Godes, D. & Silva, J.C. (2012). Sequential
parameter for an endogenous regressor. and temporal dynamics of online opinion.
Journal of Applied Econometrics, 24(3): Marketing Science, 31(3): 448-473
446-468 [43] Goh, K.Y., Heng, C.S. & Lin, Z. (2013).
[34] Fang, X., Hu, P. J.H., Li, Z. (Lionel) & Social media brand community and
Tsai, W. (2013). Predicting adoption consumer behavior: quantifying the relative
probabilities in social networks. Information impact of user- and marketer-generated
Systems Research, 24(1): 128-145 content. Information Systems Research,
[35] Fichman, R.G., Kohli, R. & Krishnan, R. 24(1): 88-107
(2011). Editorial Overview — The Role of [44] Gopinath, S., Chintagunta, P.K. &
Information Systems in Healthcare: Current Venkataraman, S. (2013). Blogs, advertising,
Research and Future Trends. Information and local-market movie box office
Systems Research, 22(3): 419-428 performance. Management Science,
[36] Flake, G.W., Lawrence, S., Giles, C.L. & forthcoming
Coetzee, F.M. (2002). Self-organization and [45] Goyal, A., Bonchi, F. & Lakshmanan, L.
identification of web communities. (2010). Learning influence probabilities in
Computer, 35(3): 66-70 social networks. Proceedings of the 3rd
[37] Ghose, A., Goldfarb, A. & Han, S.P. ACM International Conference on Web
Search and Data Mining, 2010: 241-250 Marketing Science, 30(2): 195-212
[46] Granovetter, M.S. (1973). The strength of [54] Jamali, M. & Ester, M. (2010). A matrix
weak ties. American Journal of Sociology, factorization technique with trust
1973: 1360-1380 propagation for recommendation in social
[47] Gu, B., Park, J. & Konana, P. (2012). networks. Proceedings of the 4th ACM
Research note — the impact of external Conference on Recommender Systems, 2010:
word-of-mouth sources on retailer sales of 135-142
high-involvement products. Information [55] Jing, X. & Xie, J. (2011). Group buying: a
Systems Research, 23(1): 182-196 new mechanism for selling through social
[48] Hildebrand, C., Häubl, G., Herrmann, A. & interactions. Management Science, 57(8):
Landwehr, J.R. (2013). When social media 1354-1372
can be bad for you: community feedback [56] Kempe, D., Kleinberg, J. & Tardos, É.
stifles consumer creativity and reduces (2003). Maximizing the spread of influence
satisfaction with self-designed products. through a social network. Proceedings of the
Information Systems Research, 24(1): 14-29 9th ACM SIGKDD International Conference
[49] Ho, C., Wu, J. & Tan, Y. (2013). on Knowledge Discovery and Data Mining,
Disconfirmation effect on online opinion 2003:137-146
expression. Working paper [57] Kernighan B.W. & Lin S. (1970). An
[50] Hoff, P.D., Raftery, A.E. & Handcock, efficient heuristic procedure for partitioning
M.S. (2002). Latent space approaches to graphs. Bell Systems Technical Journal, 49:
social network analysis. Journal of the 291-307
American Statistical Association, 97(460): [58] Kimura, M. & Saito, K. (2006). Tractable
1090-1098 models for information diffusion in social
[51] Hosanagar, K., Han, P. & Tan, Y. (2010). networks. Knowledge Discovery in
Diffusion models for peer-to-peer (p2p) Databases: PKDD 2006. Springer Berlin
media distribution: on the impact of Heidelberg, 2006: 259-271
decentralized, constrained supply. [59] Kleemann, F., Voß, G.G. & Rieder, K.
Information Systems Research, 21(2): (2008). Un(der)paid Innovators: the
271-287 commercial utilization of consumer work
[52] Ienco, D., Bonchi, F. & Castillo, C. (2010). through crowdsourcing. Science, Technology
The meme ranking problem: Maximizing & Innovation Studies, 4(1): 5-26
microblogging virality. IEEE International [60] Konstas, I., Stathopoulos, V. & Jose, J.M.
Conference on Data Mining Workshops, (2009). On social networks and collaborative
2010: 328-335 recommendation. Proceedings of the 32th
[53] Iyengar, R., Van den Bulte, C. & Valente, ACM SIGIR International Conference on
T.W. (2011). Opinion leadership and social Research and Development in Information
contagion in new product diffusion. Retrieval, 2009: 195-202
[61] Krackhardt, D. (1992). The strength of they keep: friendship networks and
strong ties: The importance of philos in information asymmetry in online
organizations. Networks and Organizations: peer-to-peer lending. Management Science,
Structure, Form, and Action, 1992: 216-239 59(1): 17-35
[62] Kumar, V., Bhaskaran, V., Mirchandani, R. [69] Lin, Y.R., Chi, Y., Zhu, S., Sundaram, H.
& Shah, M. (2013). Practice prize & Tseng, B.L. (2009). Analyzing
winnercreating a measurable social media communities and their evolutions in dynamic
marketing strategy: increasing the value and social networks. ACM Transactions on
ROI of intangibles and tangibles for hokey Knowledge Discovery from Data, 3(2): 8
pokey. Marketing Science, 32(2): 194-212 [70] Liu, Y. (2006). Word of mouth for movies:
[63] Kwak, H., Lee, C., Park, H. & Moon, S. its dynamics and impact on box office
(2010). What is twitter, a social network or a revenue. Journal of Marketing, 70(3): 74-89
news media? Proceedings of the 19th ACM [71] Luo, X., Zhang, J. & Duan, W. (2013).
International Conference on World Wide Social media and firm equity value.
Web, 2010: 591-600 Information Systems Research, 24(1):
[64] Lam, S. & Riedl, J. (2004). Shilling 146-163
recommender systems for fun and profit. [72] Manski, C.F. (1993). Identification of
Proceedings of the 13th International endogenous social effects: the reflection
Conference on World Wide Web, 2004: problem. The Review of Economic Studies,
309-402 60(3): 531-542.
[65] Lancichinetti, A., Fortunato, S. & Kertész, [73] Mayzlin, D. & Yoganarasimhan, H. (2012).
J. (2009). Detecting the overlapping and Link to success: how blogs build an audience
hierarchical community structure in complex by promoting rivals. Management Science
networks. New Journal of Physics, 11(3): 58(9): 1651-1668
033015 [74] Mehta, B. & Nejdl, W. (2009).
[66] Lee, M., Kim, M. & Peng, W. (2013). Unsupervised strategies for shilling detection
Consumer reviews: reviewer avatar facial and robust collaborative filtering. User
expression and review valence. Internet Modeling and User-Adapted Interaction,
Research, 23(2): 116-132 19(1-2): 65-97
[67] Leskovec, J., Krause, A., Guestrin, C. & [75] Melville, P., Gryc, W. & Lawrence, R.D.
Faloutsos, C. (2007). Cost-effective outbreak (2009). Sentiment analysis of blogs by
detection in networks. Proceedings of the combining lexical knowledge with text
13th ACM SIGKDD International classification. Proceedings of the 5th ACM
Conference on Knowledge Discovery and SIGKDD International Conference on
Data Mining, 2007: 420-429 Knowledge Discovery and Data Mining,
[68] Lin, M., Prabhala, N.R. & Viswanathan, S. 2009: 1275-1284
(2013). Judging borrowers by the company [76] Miller, A.R. & Tucker, C. (2013). Active
social media management: the case of health markets. Management Science, 58(11):
care. Information Systems Research, 24(1): 1963-1981
52-70 [85] Oh, W. & S. Jeon. (2007). Membership
[77] Mobasher, B., Burke, R., Bhaumik, R. & herding and network stability in the open
Williams, C. (2007). Toward trustworthy source community: the ising perspective.
recommender systems: an analysis of attack Management Science, 53(7): 1086-1101
models and algorithm robustness. ACM [86] Palla, G., Derényi, I., Farkas, I. & Vicsek,
Transactions on Internet Technology, 7(4): T. (2005). Uncovering the overlapping
1-40 community structure of complex networks in
[78] Moe, W.W. & Schweidel, D.A. (2012). nature and society. Nature, 435(7043):
Online product opinions: incidence, 814-818
evaluation, and evolution. Marketing Science, [87] Pandit, S., Yang, Y. & Chawla, N.V.
31(3): 372-386 (2012). Maximizing information spread
[79] Narayanam, R. & Narahari, Y. (2010). A through influence structures in social
shapley value-based approach to discover networks. Proceedings of the 12th IEEE
influential nodes in social networks. IEEE International Conference on Data Mining
Transactions on Automation Science and Workshops, 2012: 258-265
Engineering, 2010(99): 1-18 [88] Pang, B., Lee, L. & Vaithyanathan, S.
[80] Nepusz, T., Petróczi, A., Négyessy, L. & (2002). Thumbs up?: sentiment classification
Bazsó, F. (2008). Fuzzy communities and the using machine learning techniques.
concept of bridgeness in complex networks. Conference on Empirical methods in natural
Physical Review E, 77(1): 016107. language processing, 10: 79-86
[81] Newman, M.E. (2004). Fast algorithm for [89] Rabe-Hesketh, S. & Skrondal, A. (2004).
detecting community structure in networks. Generalized latent variable modeling:
Physical review E, 69(6): 066133 Multilevel, longitudinal, and structural
[82] Newman, M.E. (2006). Modularity and equation models. Boca Raton: CRC
community structure in networks. [90] Read, J. (2005). Using emoticons to reduce
Proceedings of the National Academy of dependency in machine learning techniques
Sciences, 103(23): 8577-8582 for sentiment classification. Proceedings of
[83] O’Mahony, M., Hurley, N., Kushmerick, N. the ACL Student Research Workshop,
& Silvestre, G. (2004). Collaborative 2005:43-48
recommendation: a robustness analysis. [91] Rishika, R., Kumar, A., Janakiraman, R. &
ACM Transactons on Internet Technology, Bezawada, R. (2013). The effect of
4(4): 344-377 customers’ social media participation on
[84] Oestreicher-Singer, G. & Sundararajan, A. customer visit frequency and profitability: an
(2012). The visible hand? demand effects of empirical investigation. Information Systems
recommendation networks in electronic Research, 24(1): 108-127
[92] Saez-Trumper, D., Comarela, G., Almeida, [99] Slater, P.B. (2008). Established clustering
V., Baeza-Yates, R. & Benevenuto, F. procedures for network analysis. arXiv
(2012). Finding trendsetters in information preprint arXiv:0806.4168
networks. Proceedings of the 18th ACM [100] Snijders, T.A.B. (2006). Statistical
SIGKDD International Conference on methods for network dynamics. Proceedings
Knowledge Discovery and Data Mining, of the XLIII Scientific Meeting, 2006 (1994):
2012: 1014-1022 281-296
[93] Saito, K., Kimura, M., Ohara, K. & [101] Song, X., Chi, Y., Hino, K. & Tseng, B.L.
Motoda, H. (2010). Selecting information (2007). Information flow modeling based on
diffusion models over social networks for diffusion rate for prediction and ranking.
behavioral analysis. Machine Learning and Proceedings of the 16th International
Knowledge Discovery in Databases. Conference on World Wide Web, 2007:
Springer Berlin Heidelberg, 2010: 180-195. 191-200
[94] Sasidharan, S., Santhanam, R., Brass, D.J. [102] Sonnier, G.P., McAlister, L. & Rutz, O.J.
& Sambamurthy, V. (2012). The effects of (2011) A dynamic model of the effect of
social network structure on enterprise online communications on firm sales.
systems success: a longitudinal multilevel Marketing Science, 30(4): 702-716
analysis. Information Systems Research, [103] Stonebraker, M. (2010). SQL databases v.
23(3): 658-678 NoSQL databases. Communications of the
[95] Schafer, J.B., Frankowski, D., Herlocker, J. ACM, 53(4): 10-11
& Sen, S. (2007). Collaborative filtering [104] Sun, M. (2012). How does the variance of
recommender systems. Springer Berlin product ratings matter? Management Science,
Heidelberg, 2007: 291-324 58(4): 696-707
[96] Schwienbacher, A. & Larralde, B. (2010). [105] Susarla, A., Oh, J.H. & Tan, Y. (2012).
Crowdfunding of small entrepreneurial Social networks and the diffusion of
ventures. Handbook of entrepreneurial user-generated content: evidence from
finance, forthcoming youtube. Information Systems Research,
[97] Shriver, S.K., Nair, H.S. & Hofstetter, R. 23(1): 23-41
(2013). Social ties and user-generated [106] Symeonidis, P., Nanopoulos, A. &
content: evidence from an online social Manolopoulos, Y. (2008). Providing
network. Management Science, 59(6): justifications in recommender systems. IEEE
1425-1443 Transactions on Systems, Man and
[98] Singh, P.V., Tan, Y. & Youn, N. (2011). A Cybernetics, Part A: Systems and Humans,
Hidden Markov Model of Developer 38(6): 1262-1272
Learning Dynamics in Open Source [107] Tan, P.N., Steinbach, M. & Kumar, V.
Software Projects. Information Systems (2005). Introduction to Data Mining.
Research, 22(4): 790-807 Addison-Wesley
[108] Tang, J., Sun, J., Wang, C. & Yang, Z. attack detection for securing collaborative
(2009). Social influence analysis in recommender systems. Technical Report,
large-scale networks. Proceedings of the 15th DePaul University
ACM SIGKDD International Conference on [117] Wu, L. (2013). Social network effects on
Knowledge Discovery and Data Mining, productivity and job security: evidence from
2009: 807-816 the adoption of a social networking tool.
[109] Tang, L., Liu, H. & Zhang, J. (2012). Information Systems Research, 24(1): 30-51
Identifying evolving groups in dynamic [118] Wu, Z., Cao, J., Wu, J., Wang, Y. & Liu, C.
multimode networks. IEEE Transactions on (2013). Detecting genuine communities from
Knowledge and Data Engineering, 24(1): large-scale social networks: a pattern-based
72-85 method. The Computer Journal, forthcoming
[110] Toubia, O. & Stephen, A.T. (2013). [119] Wu, Z., Wu, J., Cao, J. & Tao D. (2012).
Intrinsic vs. Image-related utility in social HySAD: a semi-supervised hybrid shilling
media: why do people contribute content to attack detector for trustworthy product
twitter? Marketing Science, 32(3): 368-392 recommendation. Proceedings of the 18th
[111] Trusov, M., Bodapati, A.V. & Bucklin, R.E. ACM SIGKDD International Conference On
(2010). Determining influential users in Knowledge Discovery And Data Mining,
internet social networks. Journal of 2012: 985-993
Marketing Research, 47(4): 643-658 [120] Yan, L., Peng, J. & Tan, Y. (2013).
[112] Wang, X., Wei, F., Liu, X., Zhou, M. & Network dynamics: how do we find patients
Zhang, M. (2011). Topic sentiment analysis like us? Working paper
in twitter: a graph-based hashtag sentiment [121] Yan, L. & Tan, Y. (2013). Feel blue so go
classification approach. Proceedings of the online: an empirical study of online social
20th ACM International Conference on support among patients. Working paper
Information and Knowledge Management, [122] Yang, T., Chi, Y., Zhu, S., Gong, Y. & Jin,
1031-1040 R. (2009). A bayesian approach toward
[113] Wasserman, S. & Faust, K. (1994). Social finding communities and their evolutions in
network analysis: methods and applications. dynamic social networks. In SDM, 2009:
Cambridge University Press 990-1001
[114] Weng, J., Lim, E.P., Jiang J. & He, Q. [123] Ye, M., Yin, P. & Lee, W.C. (2010).
(2010). Twitterrank: finding topic-sensitive Location recommendation for location-based
influential twitterers. Proceedings of the 3rd social networks. Proceedings of the 18th
ACM International Conference on Web SIGSPATIAL International Conference on
Search and Data Mining, 2010: 261-270 Advances in Geographic Information
[115] White, T. (2012). Hadoop: the definitive Systems, 2010: 458-461
guide. O'Reilly [124] Zeng, X. & Wei, L. (2013). Social ties and
[116] Williams, C. (2006). Profile injection user content generation: evidence from flickr.
Information Systems Research, 24(1): 71-87 interest in solving the problems raised from the
[125] Zhang, J. & Liu, P. (2012). Rational emerging big-data applications. His research
herding in microloan markets. Management was supported by over 20 grants from NSFC,
Science, 58(5): 892-912 MOE, MOST and MIIT. He has published one
[126] Zhang, S., Wang, R.S. & Zhang, X.S. monograph in Springer and over fifty papers in
(2007). Uncovering fuzzy community refereed conference proceedings and journals,
structure in complex networks. Physical such as KDD, SCIENCE, DMKD and TKDE.
Review E, 76(4): 046103 He is the recipient of the NSFC Excellent Young
[127] Zhang, X. & Zhu, F. (2011). Group size Scholars award (2013), the National Excellent
and incentives to contribute: a natural Doctoral Dissertation award (2010), and the
experiment at chinese wikipedia. American New Century Excellent Talents in University
Economic Review, 101(4) : 1601-1615 award (2011). He is a member of ACM, IEEE,
[128] Zhao, J., Dong, L., Wu, J. & Xu, K. (2012) INFORMS, AIS, and CCF.
Moodlens: an emoticon-based sentiment
analysis system for chinese tweets. Haoyan Sun is a doctoral student in Information
Proceedings of the 18th ACM SIGKDD Systems at the Michael G. Foster School of
International Conference on Knowledge Business, University of Washington. Her
Discovery and Data Mining, 2012: research interests include online trust, social
1528-1531 networks, and electronic commerce. She has
[129] Zhao, J., Lui, J., Towsley, D., Guan, X. & published in International Conference on
Wang, P. (2013). Social sensor placement in Information Systems.
large scale networks: a graph sampling
perspective. arXiv preprint arXiv:1305.6489 Yong Tan is the Neal and Jan Dempsey
[130] Zhao, Y., Yang, S., Narayan, V. & Zhao, Y. Professor of Information Systems at the Michael
(2013). Modeling consumer learning from G. Foster School of Business, University of
online product reviews. Marketing Science, Washington. He received his Ph.D. in Physics
32(1): 153-169 and Ph.D. in Business Administration, both from
the University of Washington. He was a
Junjie Wu received his Ph.D. degree in postdoctoral fellow at the University of
Management Science and Engineering from Strathclyde and a visiting scientist at the
Tsinghua University in 2008. He also holds a Laboratoire de Physique Quantique, Université
B.E. degree in Civil Engineering from the same Paul Sabatier. His research interests include
university. He is currently an Associate professor electronic, mobile and social commerce,
in Information Systems Department, School of economics of information systems, social and
Economics and Management, Beihang economic networks, and software engineering.
University. His general area of research is data He has published in Management Science,
mining and complex networks, with a special Information Systems Research, Operations
Research, Management Information Systems Research, and is an associate editor of

Quarterly, Journal of Management Information Management Science and a senior editor of
Systems, INFORMS Journal on Computing, Journal of Electronic Commerce Research. He
IEEE/ACM Transactions on Networking, IEEE served as a co-chair of Conference on
Transactions on Software Engineering, IEEE Information Systems and Technology (CIST
Transactions on Knowledge and Data 2010) and the cluster chair of 2012 INFORMS
Engineering, IIE Transactions, European Journal Information Systems Cluster, and is a track chair
on Operations Research, Decision Support of International Conference on Information
Systems, among others. He served as an Systems (ICIS 2013).
associate editor of Information Systems

Wu2013 Article SocialMediaResearchAReview

Uploaded by

Copyright:

Available Formats

Wu2013 Article SocialMediaResearchAReview

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wu2013 Article SocialMediaResearchAReview

Uploaded by

Copyright:

Available Formats

J Syst Sci Syst Eng (Sep 2013) 22(3): 257-282 ISSN: 1004-3756 (Paper) 1861-9576 (Online)

DOI: 10.1007/s11518-013-5225-6 CN11-2983/N

SOCIAL MEDIA RESEARCH: A REVIEW

Junjie WU1 Haoyan SUN2 Yong TAN2

1. Introduction Different social media platforms play

 Systems Engineering Society of China and Springer-Verlag Berlin Heidelberg 2013

designation and find that although opinion movie performances.

4. Predictive Analytics social media mining research in some subfields

Conference On Computer Supported (2012). How is the mobile internet different?

Research, Management Information Systems Research, and is an associate editor of

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.