National Taiwan University of Science and Technology D10008104@mail.ntust.edu.tw
This research was sponsored by the National Science Council of Taiwan, under the project number NSC 101-2410-H-011 -002 -
Network effect: Information diffusion in social networking site
Abstract
Social website has become part of our lifes for us to spread information and connect each others, the huge online social networks becomes a powerful recommendation and interaction mechanism system. This study designed a Facebook application to uncover the effects of individuals’ connection patterns on information diffusion process on Facebook. The results showed that both degree and clustering effect have significant impact to information dissemination process on Facebook. That is, the one with more connections and higher network cluster members will has more readers during the dissemination process and the information has higher possibility to retransmit to the network. The findings of this study provided useful implications not only for theory in the network effect, but also useful references and suggestions to the marketing practitioners.
Keywords
Social website, information diffusion, Facebook, network effect, clustering effect
Introduction
People use social website to perform many different function-marketing tools, connect with people, form revolutions and more.
Social website has become part of our lifes for us to pass on information and connect eachothers. The social website is a tool for people to communicate and maintain relationships, therefore, the user unit is individual with their actual social network as foundation and further develop online social network (Jahan and Ahmed, 2012). With the development of SNS, SNS has now become a tool for organization or event holder to promote (Arora, 2012). The information distribution pattern of SNS is different from the traditional broadcasting which is basically communicating information to as many people as possible with no need to know the receivers. SNS, on the other hand, distribute information on the individual network basis; the effect of the information distribution in the social network shall be affected by the individual's power of effect in the social network (Jahan and Ahmed, 2012).
The emergence of social network sites (e.g., Facebook, Twitter) has the potential to dramatically alter an individual’s exposure to new information (Bakshy et al., 2012; Wilson, Fornasier and White, 2010). It changed the form of communication into collaboration and sharing, and bring the idea of tribe into community and becoming a deliberative community with diversity, which enables individual to maintain and develop relationships with each other (Jahan and Ahmed, 2012; Wan and Yu, 2012; Zhou, 2012).
Besides, social network sites also provide functions for their users to share information within their social networks easily (Jahan and Ahmed, 2012; Liang et al., 2011). Facebook’s “Share” button and Twitter’s “Tweet” button are samples of this feature (Harrigan, Achananuparp and Lim, 2012). With this feature, people can share the news with all of their friends immediately, and instantly receive responses from others who are interested. In this thread, the huge online social networks may become a powerful recommendation and interaction mechanism system (Arora, 2012; Hinson, 2011).
The primary objective in this study was to uncover the effects of individuals’ connection patterns on information diffusion process in online social networks. Network effects, relates to the influence of the structure of connection patterns, which can distinguish in degree effect and clustering effect (Katona, Zubcsek and Sarvary, 2011). Degree means the number of connections an individual has. Katona, Zubcsek and Sarvary (2011), point out the more connections also brings the higher degree. The development of SNS greatly increases the degree of one’s weak ties and support on forming and maintaining weak ties, because the technology is well-suited to maintaining such ties cheaply and easily (Donath and Boyd, 2004). Thus, this study supposed that when a user has more connections on SNS than others have, they have more chance to receive more information than others and has higher
possibility to spread information to more people. In addition to the number of connections, the density of connections in a group may also affect the individual on receiving information. Watts (2002) claimed that denser networks have a stronger influence on its member and more possible paths for information to travel, which makes information flows easier. Thus, this study also suggested that the higher the network cluster, the more easily to disseminate information in the network.
The appearance of Facebook provides a way to evaluate the path of information dissemination which is difficult in the past. Via the connecting of Facebook, users can examine the record of interaction between people more easily and publicly, and further build up personal network topology (Lewis et al., 2008). Facebook record the social interaction completely and systematic which was difficult to be recorded, and these information are helpful on analyzing interpersonal relationships (Petroczi, Nepusz and Bazso, 2007). Therefore, this study take Facebook as a SNS sample, and simulates an information-spreading process via building a program to record the situation and timing of dissemination, disseminators and receivers. This study collected the related data of dissemination process to investigate the correlation between disseminator’s influential power in information spreading and disseminator’s network effect. The results are able to understand how social networks can be used to provide better target marketing strategies.
Theoretical Background
The information is disseminated on SNS
Boyd and Ellison (2007) define SNSs as web-based services, which allow individuals to (1) construct a public or semi-public profile within a bounded system, (2) articulate a list of other users with whom they share a connection, and (3) view and traverse their list of connections and those made by others within the system. Based on these three functions, the users of SNSs can connect with whom they wanted to and eventually structure SNSs with their personal networks. This more accurately mirrors unmediated social structures, where “the world is composed of networks, not groups” (Ellison et al., 2007; Weimann, 1991).
Structural variations around visibility and access are the primary ways that SNSs differentiate themselves from each other;
SNSs are variations in the information and communication tools, which incorporate both social and technological forces caused users to have different practice and behavior in each SNSs (Boyd and Ellison 2007). Scholars have observed SNSs by features and user base, an SNS can either be anonymous or nonymous, serve diverse audiences or target specific demographics, designed for meeting new people or maintain pre-existing social connections (Boyd and Ellison, 2007; Ellison, Steinfield and Lampe, 2007; Ross et al., 2009). They represent rich and popular communication interfaces for hundreds of millions of users. Users can exhibit their demographics as well as their preferences by carefully editing their profiles, and reveal their likely communication patterns (Katona, Zubcsek and Sarvary, 2011).
Facebook, the first rank SNS, is considered to be the third largest country in the world. It is a social communication tool designed to allow users to contact and communicate with other users (Nosko, Wood and Molema, 2010; Ross et al, 2009), which is nonymous, maintain pre-existing social connections and diverse audiences, but its function can also serve user to meet new people with common interest. Users can convey desirable information about themselves (via features such as About Me, Notes, and Status Updates routinely found on SNS) and can select attractive, self-promoting photographs (Mehdizadeh, 2010). For instance, people can send messages to their friends, share their pictures, and maintain a profile page with personal information. They can also join virtual groups to interact with others based on common interests (Ellison, Steinfield and Lampe, 2007; Hinson, 2011; Pempek, Yermolayeva and Calvert, 2009). Besides the basic message and comment interaction, Facebook also provides other easy access and useful tools for users to spread information, such as like, share, or tag (Liang et al., 2011; Pempek, Yermolayeva and Calvet, 2009).
The main purpose of social network sites is sharing information with friends (Hinson, 2011; Liang et al., 2011). Boyd and
Ellison (2007) and Taylor, Lewin and Strutton (2011) also suggested that Facebook’s friends are mostly came from pre-existing social connections, that can cause individual to consider the information from Facebook is more reliable than other commercial message and anonymous WOM. Also, the interconnection between nodes in social network enhances the process of information dissemination and amplifies the influence within each other. Therefore, this study discussed the influence of network effect during the information dissemination, based on network structure point of view.
Network effect
Network effects, relates to the influence of the structure of connection patterns (Katona, Zubcsek and Sarvary, 2011). Brown and Reingen (1987), and Vilpponen, Winter and Sundqvist (2006) confirmed that individuals’ WOM behavior is influenced by the properties of social relationships among them, such as network structure. It can be distinguished to degree effect and clustering effect.
The relationship between degree effect and information disseminate
Degree means the number of connections an individual has (Katona, Zubcsek and Sarvary, 2011). It is the probability of the number of links per node has over the entire network (Dover, Goldenberg and Shapira, 2012).
Gonzales, Hancock and Pennebaker (2010) claimed that with Facebook visualizing and displaying the friends’ connections, a large number of Facebook friends could remind users of their social connections and boost their self-esteem, which could enhance their subjective well-being level. Besides, because the technology is well-suited to maintaining weak ties cheaply and easily, SNSs could greatly increase on forming and maintaining weak ties (Donath and Boyd, 2004). These weak ties is a bridge social capital that allowing users to create and maintain larger, diffused networks of relationships from which they could potentially draw resources (Donath and Boyd, 2004). Lim and Putnam (2010) and Granovetter (1982) also claimed that loose connections between individuals may provide useful information and new perspectives for one another but typically less on emotional support. Thus, this study suggested that the degree will significant impact information disseminate on Facebook. Based on prior works, this study proposes the following hypothesis:
H1: As network degree increases, the level of information dissemination will be wider.
Furthermore, Krackhardt (1998) claimed that to evaluate the influence of individuals’ contacts should not only count the number of related actors but also need to examine how those relationships are embedded in the entire network of relationships. This study shall discuss how clustering impact the information disseminate on Facebook.
The relationship between clustering effect and information disseminate
Clustering characterizes the density of connections in the network. Dense network means a closed or integrated network, in which the alters know each other well and the dilute network is an open or radial network where the alters hardly know each other (Sohn, 2009). Girvan and Newman (2002) also suggested that denser networks have more connections and are usually close-knit and well-defined communities, which is benefit for a seed because in-group members are likely to help and reward each other. Thus, network density does not indicate how close the ego and the alters are in the network, but reveal which the nodes are interconnected (Scott, 2000). Clustering coefficient is used to measure the extent to which a set of members are interconnected (Watts and Strogatz, 1998). It is clear that this measure might be relevant in a context in which these members exert an influence on another member.
Network closure theory proposes that when two related individuals are connected to the same third parties, the network becomes more effective at transmitting information, and the affected relationships ultimately become stronger (Burt, 2005; Coleman, 1988). The shared third parties create redundant paths for information flow, leading to increased trust between the two related actors.
Moreover, Katona, Zubcsek and Sarvary (2011) also suggested that a set of highly connected individuals has a stronger influence on a potential adopter than an identical number of sparsely connected ones.
In general, the density of network can be used to describe how easily information can possibly flow or spread throughout the network (Katona, Zubcsek and Sarvary, 2011; Stephen et al., 2012). Watts (2002) also suggested that denser networks have more possible paths for information to travel, which therefore makes information flows easier. Thus, this study suggests that the closer the network cluster (density), the easily it is to disseminate information on Facebook. And then proposes the following hypothesis:
H2: As network cluster increases, the efficiency of information dissemination will be higher.
Moreover, Katona, Zubcsek and Sarvary (2011) also suggested that when keeping the density of one’s personal network constant, the number of relationships in the network increases contact with the number of related actors. In other words, when a larger personal network with the same clustering coefficient has more ties per neighbor between its members, they may have high information disseminates. This study also suggested this situation on Facebook, and then proposed the following hypothesis:
H3: As network cluster constant, the higher the network degree is, the higher efficiency of information dissemination will be.
Method Procedure
To observe the information flow in Facebook, this study designed a Facebook application to examine how user spread information on social networking sites. The experiment was taken place in a 7-day period in 2012. All users whom received or transmitted the information did not know the whole information dissemination process was an experiment. After the information dissemination process ended, this study recorded and observed the extent of dissemination by the number of nodes in the network that had retransmitted the information. These records were helpful to draw the information spreading path on Facebook, and further clarified the influence of transmitters’ social network structure to information dissemination (see Figure 1).
This study selected a heavy Facebook user as initial seed transmitter, the user has more than a thousand connections and subscriber on Facebook, at least three hours of active using per day and regularly posting and commenting on Facebook. To start the information dissemination, this initial seed transmitter posted the designed information with the link of Facebook application on Facebook wall. When other users used the application and decided to retransmit it, the application will post a pre-designed message on their Facebook wall. Then, this study collected the information from disseminator to calculate the network degree and clustering coefficient. The results were contributed to examine the diffusion of online social networking sites.
Figure 1. Procedure of the data collection this study
Create Facebook
Development tool and Material
Tool
The program was developed to operate on Microsoft Windows Server 2008 R2 Standard, Apache and MySQL which provided by AppServ 2.5.9 installer package, use PHP and JavaScript as programming language. For accessing the social network dataset on Facebook, the program used Facebook SDK and the Facebook Query Language (FQL) object to connect participant’s Facebook account and access their basic information and interaction records on Facebook.
To require the information that can reveal user’s characteristics on Facebook, the program needs to access their basic information and interaction records on Facebook with permission. Some of interaction records have an available period that Facebook allow to access, basic on this limitation we need to carefully decide what type of interaction is useful for this study. To understand the status of each disseminator’s social network structure, this study collected the number of Facebook friends, the number of mutual friends between each connection, the frequency of posting, and the level of profile exposing. Besides, this program records each spread point in the 7-day period and the information disseminator and receiver (see Figure 2).
Figure 2. Outline of the data acquisition process
Designed Information
In order to attract users to spread the information and grant permission, this study designed a persona test application as the object to disseminate. Persona test usually required users personal information and the content can be designed with interesting
User read the message from friend
User decided to use the program
Request to grant permission for program
Save permission to database Yes
Program authorized?
Yes
No
Show the pre-designed message to user
User wants to spread the program?
Yes Post message on user’s
wall
End
pictures and text that is helpful on increase users’ willing of grant permission. The property of content will affect the type of transmitter and limit the dissemination. Thus, this study designed a neutral persona test with comment of personality and the picture of a well-known person who has the same personality to avoid deviation (see Figure 3). In the persona test, there are 36 different results with relative comments and pictures that will be provided base on user’s date of birth. In the end, user can see their own result after granted permission, and user can choose to post it on Facebook wall or not.
Figure 3. Designed information in this study Transmission
In according with Watts and Dodds (2007) inferred, this study also assumed that individual i in a population of size N influences ni others, where ni is drawn from an influence distribution p(n). The single node is selected as the initial transmitter at time t=0. This node transmits a piece of information to all ni friends on Facebook. The ni exposed nodes each independently decide whether to view the information with probability q = P (view│exposure). And then exposed nodes that viewed the information decide whether to retransmit to their friends. Thus, ni are the part of the population of size N who has been influenced by particular information.
First, the dissemination proceeds from an initial state in which all individual are inactive (state 0), then chosen initial i who is activated (exogenously) to transmit the information on his newsfeed. This study expected the information will newly activated his friends, and then continue trigger their friends like chain reaction, which is generating a sequence of activations. This is what Watts (2002) called cascade. Thus, when all activations associated with a single cascade, its size can be determined simply as the total cumulative number of activations. The program we built will record the basic personal information of transmitter and the timing of dissemination, which are helpful to draw the information spreading path on Facebook for further observing the quantity and the level of dissemination of each individual transmitter. The ending of information dissemination usually happened when the receiver has no willingness to retransmit, and the receiver’s friend will not be able to know the information.
This study used the software, Pajek, to draw the path of information dissemination (see Figure 4). Pajek is a program for analysis and visualization of large networks having tens or hundreds of thousands of vertices. In the drawing process, this study first filter global dissemination by the number of nodes each path has and excludes those which has less than three nodes as the dissemination effect is too low. After the filtering step, we examine each local dissemination individual (see Figure 4) to clarify each disseminator’s consequent of information spreading on Facebook. The results found out that some individuals has a lot of people read and retransmit the information they spread, and some has only a few.
Figure 4. the path of local information dissemination Measure
Degree (Di). The degree Di is the number of connections that node i has with other nodes in the network (Boccaletti et al., 2006).
The primary dataset for this work was collected through an application that was created on Facebook, and the numbers of their friends are readily available and easily accessible.
Clustering coefficient (Ci). Clustering coefficient measures the extent to which a set of members are interconnected (a local property) (Watts and Strogatz, 1998). It was cleared that this measure might be relevant in the context in which these members exert an influence on another member. More specifically, Ci is defined as the ratio between the number of connections ei among the ki
neighbors of node i and the maximum possible number of connections between these neighbors which is ki (ki-1)/2 (Watts and Strogatz, 1998; Whitacre, Sarker and Pham, 2011). Thus, ki = (node i and mutual friends with their friends j +1). l means the virtual connection between i and j. (ki-1) presenting the number of mutual friends between i and j, which means in the network of i all the nodes besides i itself, the triplet number = ( the number of mutual friends + 1) multiplied by the number of mutual friends / 2. The number of triangles = the number of mutual friends. The clustering coefficient for a network c is simply the average of ci value.
neighbors of node i and the maximum possible number of connections between these neighbors which is ki (ki-1)/2 (Watts and Strogatz, 1998; Whitacre, Sarker and Pham, 2011). Thus, ki = (node i and mutual friends with their friends j +1). l means the virtual connection between i and j. (ki-1) presenting the number of mutual friends between i and j, which means in the network of i all the nodes besides i itself, the triplet number = ( the number of mutual friends + 1) multiplied by the number of mutual friends / 2. The number of triangles = the number of mutual friends. The clustering coefficient for a network c is simply the average of ci value.