Scholarly article on topic 'Leveraging the Power of a Twitter Network for Library Promotion'

Leveraging the Power of a Twitter Network for Library Promotion Academic research paper on "Educational sciences"

CC BY-NC-ND
0
0
Share paper
OECD Field of science
Keywords
{Twitter / "Social media" / Networks / Influence / Metrics / Follower}

Abstract of research paper on Educational sciences, author of scientific article — Jason Shulman, Jewelry Yep, Daniel Tomé

Abstract The Twitter network of two academic libraries was analyzed to determine the influential accounts that connect to them. Such information can be exploited by libraries to create tailored social media outreach and information dissemination programs. Three network metrics, measuring different definitions of importance, were calculated for each account in the network. This allowed for the quantification and ranking of the accounts by influence/importance, normally considered to be qualitative and subjective. By all measures, accounts associated with the institutions, and not faculty, staff, or students, were found to be the most influential players in the networks of both libraries, suggesting that this is a general feature of academic library Twitter networks. Furthermore, the library, as an institutional account itself, is also influential to the broader Twitter community of its home institution. This demonstrates that the library is in a key position to propagate information from sister accounts at the institution.

Academic research paper on topic "Leveraging the Power of a Twitter Network for Library Promotion"

ELSEVIER

Contents lists available at ScienceDirect

The Journal of Academic Librarianship

Journal. Academic Librarianship ,

Leveraging the Power of a Twitter Network for Library Promotion

Jason Shulman a'*, Jewelry Yep b, Daniel Toméc

a Department of Physics, Richard Stockton College of New Jersey, 101 Vera King Farris Dr., Galloway, NJ 08205, USA b School ofHealth Sciences, Richard Stockton College ofNew Jersey, 101 Vera King Farris Dr., Galloway, NJ 08205, USA c Office of Service Learning, Richard Stockton College of New Jersey, 101 Vera King Farris Dr., Galloway, NJ 08205, USA

ARTICLE INFO

ABSTRACT

Article history: Received 8 October 2014 Accepted 16 December 2014 Available online 30 January 2015

Keywords:

Twitter

Social media

Networks

Influence

Metrics

Follower

The Twitter network of two academic libraries was analyzed to determine the influential accounts that connect to them. Such information can be exploited by libraries to create tailored social media outreach and information dissemination programs. Three network metrics, measuring different definitions of importance, were calculated for each account in the network. This allowed for the quantification and ranking of the accounts by influence/ importance, normally considered to be qualitative and subjective. By all measures, accounts associated with the institutions, and not faculty, staff, or students, were found to be the most influential players in the networks of both libraries, suggesting that this is a general feature of academic library Twitter networks. Furthermore, the library, as an institutional account itself, is also influential to the broader Twitter community of its home institution. This demonstrates that the library is in a key position to propagate information from sister accounts at the institution.

© 2014 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

INTRODUCTION

Academic libraries, in an effort to provide convenient and effective service, have been quick to implement new technologies, evaluate their efficacy, and refine their use. Perhaps the most significant recent example of this has been the incorporation of social networking into promotion and outreach activities. In particular, Twitter has become one of the most widely adopted and studied platforms. Libraries have used Twitter primarily for marketing of services and programs (Del Bosque, Leif, & Skarl, 2012; Gunton & Davis, 2012; Milstein, 2009). Many aspects of a library's Twitter program affect its effectiveness; however, they can be divided into two primary categories, properties of the tweets and properties of the associated accounts (Petrovic, Osborne, & Lavrenko, 2011; Suh, Hong, Pirolli, & Chi, 2010; Yang & Counts, 2010). Milstein (2009) and Cole (2009) present best practices involving the former. This work focuses on the latter.

Information dissemination is the primary goal of any Twitter program run by the library. In analyzing the success of such endeavors, we often simply look to the number of followers of the account. It is certainly true that the information reaches more people as the number of followers increases; however, relying solely on the number of followers to gauge the impact of a Twitter program ignores much of what makes social media an effective vehicle for outreach and dissemination of information, the network. Indeed, Twitter accounts form a network. Accounts follow and/or are followed by the library, but these accounts

* Corresponding author. E-mail address: jason.shulman@stockton.edu (J. Shulman).

can follow one another and even others not directly connected to the library. Links between accounts that are not directly connected to the library can have significant impact on information dissemination. If properly harnessed, these links can help spread the library's message well beyond its direct followers (Yep & Shulman, 2014).

This work presents an analysis of the follower/followee networks of the libraries from two primarily undergraduate state institutions, The Richard Stockton College of New Jersey and California State University San Marcos, with the goal of identifying the influential accounts connected to the library. Such information can allow those in charge of a library's account to tailor their tweeting and increase the reach and effectiveness of Twitter activities. Three metrics related to an account's influence were examined. Interestingly, it was found that the most influential accounts, by all measures, were not the students nor the faculty, but the other accounts associated with the institutions. Similar results from the analyses of both schools suggest that this is a general feature of such networks. The consequences of this finding will be detailed below. Importantly, the procedure employed in this study can be easily implemented by librarians to identify specific influential accounts at their institutions, detect communities within the network, and tailor their Twitter activities to maximize information dissemination.

A story reported by Harold Glazer (2009) of Rutgers University nicely illustrates how a single social media connection can enhance a library's electronic outreach efforts. Glazer describes the implementation of Facebook at his library. Early in the program, he noticed a marked increase in library related articles in the school's student newspaper, several of which were featured on the front page. Glazer traced the source of this new attention to the editor of the paper, who had

http://dx.doi.org/10.1016/j.acalib.2014.12.004

0099-1333/© 2014 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/40/).

connected to Glazer and the library on Facebook. The moral of this story, for the present work, is that not all social media connections are equal. As much as we might not like to admit it, some accounts wield more influence than others. The purpose of this study is to identify the influential players in the Twitter network so that they may be recruited for dissemination purposes. It also allows for the quantification of an account's influence, something that is normally qualitative. Finally, the procedure outlined below can be automated by a computer and requires few resources from library staff.

LITERATURE REVIEW

SOCIAL MEDIA USE BY PATRONS AND LIBRARIES

Two comprehensive reports were presented by Cassidy and coauthors (2011, 2014) on student technology use at Sam Houston State University. The most recent study found that more than 80% of students had a smartphone, which has led to a larger demand for mobile library services. The reports also show that student adoption of Twitter has doubled, from 21.2% to 41.4%, during the intervening years. Sixty percent of students in the 2014 study were not interested in library services using Twitter; however, the majority of Twitter users did indicate desire to connect with the library via Twitter.

Park (2010) examined the use of the popular Korean social networking site Cyworld by students, both undergraduate and graduate, and faculty. He found undergraduates open to adopting new technologies and interested in exposure and ability for self-expression afforded to them by Cyworld. In contrast, graduate students and faculty preferred to use the internet for information gathering purposes. Faculty demonstrated little social media use, while graduate students employed it for assistance with schoolwork and career advancement. Park concludes that efforts by libraries to engage faculty via social media should highlight the benefits of its use to social networking and communication.

In fact, Dickson and Holley (2010) suggest that academic libraries must advertise their social media services. As mentioned previously, Rutgers's library benefitted from the exposure their Facebook page received in the student newspaper. Perhaps one of the more interesting examples of promotion is the use of a Chinese microblogging site to successfully promote a social media marketing campaign (Luo, Wang, & Han, 2013). It's worth noting that the procedure presented below can help librarians market the account and their services by identifying influential Twitter accounts which can be employed to increase the library's exposure.

TWITTER USE BY LIBRARIES

Twitter is most commonly used to broadcast information about the library (Gunton & Davis, 2012; Milstein, 2009). Eleven out of twenty suggestions provided by Cole (2009) involve sending out information to followers, while five are devoted to promotion of the library and its services. In their broad review of Twitter use by academic libraries, Del Bosque et al. (2012) found that libraries primarily used their accounts for discussing resources (55% of libraries), announcing events (24%), and communicating hours (14%).

While Twitter is an effective tool for broadcasting library information, the scholarly community appears to be united in the belief that Twitter should be used by academic libraries to interact with followers in order to take full advantage of the platform (Cole, 2009; Cuddy, Graham, & Morton-Owens, 2010; Del Bosque et al., 2012; Dickson & Holley, 2010; Gunton & Davis, 2012; Milstein, 2009; Sewell, 2013). Gunton and Davis (2012) suggest that limiting a library's social media activities to the distribution ofinformation represents a failure to appreciate the interactive nature of the modern internet. It appears that academic libraries have been slow to adopt such advice. Del Bosque et al. (2012) found that 54% of libraries interacted with followers and less

than 10% used Twitter to answer reference questions, although most libraries did have some version of electronic reference. The authors encourage libraries to take advantage of Twitter-specific features such as direct messages, @replies (replying to an account by using @ followed by the account name), and hashtags (keywords preceded by the # symbol, e.g. #LibraryScience). Hashtags allow Twitter users to easily follow discussions on a topic simply by searching for the hashtag. Del Bosque et al. also advocate the use of library-specific hashtags.

INFORMATION DISSEMINATION VIA TWITTER

It is clear that social media is used by academic libraries as a means to transmit information to and connect with patrons. Thus, an effective program will reach many people. This is a question of information dissemination. Dissemination via Twitter has been studied by several groups. Much of the focus has been on retweeting, or the forwarding of a tweet composed by another, since, according to Suh et al. (2010), "Retweeting is the key mechanism for information diffusion in Twitter." Several account properties are associated with high levels of retweeting, the most obvious of which is the number of followers (Petrovic et al., 2011; Suh et al., 2010). Perhaps more surprisingly, the number of followees is also related to the generation of retweets (Suh et al., 2010) as is the account being a member of Twitter lists (Petrovic et al., 2011).

Kim, Abels, and Yang (2012) studied information dissemination by academic libraries. The study examined the account types that retweeted library content. Retweeters were grouped into twelve categories, e.g. librarians, students, scholars, and university organizations. University organizations did the most retweeting and was the largest intermediary, garnering more retweets of library messages than any other group. Such accounts are central to the Twitter network and are responsible for circulating much library content. Students were found to retweet messages both directly from the library and also those that were previously retweeted by other groups. A moderate amountof retweeters were categorized as local organizations; however, this group was responsible for spreading the second highest number of messages. This indicates the importance of cultivating relationships between academic libraries and such accounts.

This work is concerned with the Twitter networks and influential accounts within them. However, for completeness, it is worth noting properties of individual tweets that are associated with retweets and information propagation. The inclusion of both URLs and hashtags has been shown to increase the retweetability of a message (Petrovic et al., 2011; Suh etal., 2010; Yang & Counts, 2010). Suh et al. (2010) found that the impact of including URLs in a tweet is domain specific. Some URLs enhance the appeal of the message while others reduce it. This was also found for hashtags. Overall, however, messages with these features are more likely to be forwarded by other users.

ANALYSIS OF FOLLOWERS

Sewell (2013) performed a comprehensive investigation of the Twitter followers from a library at Texas A&M University. Such an analysis is critical to the development of targeted marketing and relevant tweets (Cuddy et al., 2010). For example, this knowledge can indicate which communities are heavily represented, and moderators can tailor content to these groups. Furthermore, once known, underrepre-sented populations can be engaged and recruited.

There were 432 accounts following the Texas A&M library's Twitter account. Each was individually examined and placed into one of eleven categories, e.g. student, faculty/staff, and alumni. Other social media sites were examined to locate this information, if necessary. In addition, Sewell created other account properties such as affiliated/unaffiliated with the University, active/inactive, and number of tweets. Categories also had subproperties associated with them. For example, student accounts were further partitioned by academic program and year.

Fig. 1. A sample follower/followee Twitter network graph. Arrows' direction indicates attention paid by one account to another. They begin and end on accounts that follow and are followed, respectively. Information, in the form of tweets, flows against the arrow direction.

Forty-five percent of the followers were associated with the University. Students represented the largest population of followers (24%). Of these, 81% were undergraduates. Corporations (20%) were the second most represented population. Other groups to note were University departments/organizations (9%) and faculty/staff (5%). Cuddy et al. (2010) found a similarly diverse set of followers in the early days of the NYU Health Sciences Libraries' Twitter account.

As mentioned above, an analysis of an account's followers can provide valuable insight into the accounts that receive library information. It can help librarians determine policies ranging from simple tweet development to large-scale policy decisions such as whether to allocate resources to social media (Sewell, 2013). However, there are difficulties associated with such endeavors, the foremost of which is the enormous time investment. The account of each follower must be visited and details copied. Personal information, such as graduating year, department, or even gender is not always easily accessible. The account must be combed if such information is needed, requiring more time. Sewell was forced to search for and access other social media accounts to locate the information required by her study. Another difficulty with a follower examination is the analysis of such a large volume of irregular data. Developing a big picture from the many details of accounts is not straightforward. The purpose of this work is to complement this microscopic analysis of a library's Twitter community with a multiscale examination. A simple method to extract communities as well as key accounts in the Twitter network is presented. Importantly, the data collection and analysis is automated, thus requiring little time from librarians.

INTRODUCTION TO NETWORKS

The field of networks is vast and continues to grow. This section will introduce only that which is relevant to the present study. More details can be found in the works of Newman (2010) and Easley and Kleinberg (2010).

TERMINOLOGY

A network is a collection ofentities that connect or interact in some manner. This definition is quite general, which allows the concept to be applied to diverse situations. Consider the following examples. In a friendship network, the entities are people and the connections are friendships. The power grid is also a network. The entities are power plants, businesses and homes, and the connections are power lines. Citation networks are formed from articles and books that cite other written works.

These examples demonstrate that the concept of networks has permeated many fields. As a consequence, redundant terminology has developed. The entities in a network are commonly referred to as vertices but are also known as nodes. The connections are called edges, interactions, or ties, to name a few. Networks can be visualized by drawing a network graph (see Fig. 1). Vertices are represented as dots (or similar) and edges are drawn as lines connecting them. In some networks, a direction can be associated with the edges. For example, in a follower/followee Twitter network, an account can follow

another, but that does not mean that the attention is reciprocated. Fig. 1 shows an example of such network. Alice follows Bob, as indicated by an arrow running from her to Bob. An arrow also points from Bob to Alice, demonstrating that he follows her too. Claire follows Dan; however, the interaction is not reciprocated. Networks in which edges have a direction are referred to as directed networks. In a follower/followee Twitter network, attention follows the direction of the arrows. Claire pays attention to Dan's tweets. Bob pays attention to Claire, while Alice pays attention to Bob who, in turn, pays attention to her. Note, however, that information flows against the arrows, in the opposite direction of attention.

METRICS

One might imagine that vertices in a Twitter network can possess differing levels of influence; some are more important in the network than others. Celebrities, for example, are retweeted much more often than others, making them sources for information and allowing them to wield considerable influence (Petrovic et al., 2011, p. 587). The importance of a vertex can be defined in many ways. Perhaps it is popular, possessing many connections. Vertices that act as a conduit for information can also be considered important, even if they do not have many connections. Sociologists have developed methods to identify central vertices in a network and rank them according to these different definitions of importance. The measures of importance are called centralities. This work discusses a few of them.

DEGREE

Degree, sometimes called degree centrality, is the number of edges attached to a vertex. In a friendship network, for example, a person's degree would be his or her total number of friends. Thus, degree can be considered a rudimentary measure of popularity (Hansen, Shneiderman, & Smith, 2010, p. 40). For directed networks, such as the Twitter follower/followee considered here, there are two degree measures, in-degree and out-degree, describing the incoming and outgoing connections, respectively. The in-degree is a measure of popularity, a measure of the number of followers within the network. The out-degree is a measure of how much attention an account pays to the others in the network. In Fig. 1, Alice has an in-degree of one and an out-degree of one. Dan has an in-degree of one and an out-degree of zero; Claire follows him, but he does not follow anyone in the network.

EIGENVECTOR CENTRALITY

There is a limitation to using a vertex's degree to quantify its significance in the network. Each connection is valued equally, so forming a connection with an important vertex counts as much as a connection to an unimportant one. It can be argued, however, that developing a connection with the head of one's company bestows more influence than a connection with someone in an entry level position. Eigenvector centrality1 accounts for this discrepancy. Like degree, the value of the

1 Eigenvector is a mathematical term that refers to the method used to calculate the rankings.

Fig. 2. An example network. The degree of each vertex is included below it Betweenness centrality is included above. Vertex E connects the left and right sides of the network. It lies on more shortest paths between its fellow vertices than any other. Consequently, it has the largest betweenness centrality. It does not, however, have the largest degree.

centrality can be large by having many connections. It can also be large if the vertex has important neighbors (Newman, 2010). In a social network, a person with many friends is influential. Eigenvector centrality accounts for the fact that a person can be influential with only a few friends who happen to be influential themselves.

BETWEENNESS CENTRALITY

Imagine a piece of information flowing between two accounts in a Twitter network. Likely, there will be multiple paths that connect the two accounts. Consider two paths, one long and one short. The first account generates the information with a tweet, and those following the account have the option to retweet it. In fact, each account that receives the information along the chain has the option to pass it along or simply ignore it. If the path is long, with many accounts, the likelihood that all accounts will pass along the information is low; often someone will

ignore it, and the information flow will die en route to the account at the end of the chain. Tweets rarely make it five steps away from the source account (Yang & Counts, 2010, p. 357). Thus, short paths tend to be important in passing along information and, consequently, accounts that lie on many short paths have considerable control over information diffusion in the network. This property is captured in a metric called betweenness centrality. Vertices with a high betweenness centrality lie on many of the shortest paths between the other vertices in the network. They act as gatekeepers of information.

Fig. 2 shows a network in which each vertex is decorated with its degree (below) and betweenness centrality (above). Vertex E acts as a bridge connected the clusters on the left and right side of the graph. It has the highest value of betweenness centrality because any information flowing from the left cluster to the right (and vice versa) must pass through vertex E. In fact, if E was removed from the network,

Fig. 3. The RSC Library network. Vertex and label size are proportional to betweenness centrality. Vertex color reflects eigenvector centrality, with larger values corresponding to darker vertices. Labels for accounts owned by individuals are not included.

there could be no communication between the left and the right sides. Clearly vertex E is important to the network. Note, however, that it has the lowest degree (a tie). This example underscores the need to examine several metrics. By considering only the degree of the vertices, one would rule out E as a significant player in the network. Vertices A, E, and I have the same degree; however, A and I have the lowest betweenness centrality. Unlike E, their removal would not impede information flow between the remaining vertices. Information could start at any vertex and reach any other, even if one or both are removed.

METHODOLOGY

NodeXL2 was used to both construct and analyze the follower/ followee Twitter networks of the libraries. It was chosen for several reasons. First, it is a freely available add-on to the ubiquitous Microsoft Excel spreadsheet software. As such, familiarity with Excel translates into proficiency with NodeXL. More importantly, however, is NodeXL's automatic download of Twitter, and other social media, data. It also has built in network analysis tools. Therefore, access to and analysis of social media data can be accomplished in a matter of a few clicks of a mouse without requiring any programming knowledge from the user. This permitted the simple workflow used in this study, which is summarized below.

1. Download data. NodeXL was used to download the libraries' follower/ followee networks from Twitter. Twitter limits the rate at which data can be accessed. It took 1 -2 days to download the complete networks for each library. Data from the Richard Bjork Library at The Richard Stockton College of New Jersey was saved on March 27, 2014. The library's Twitter handle is @RSC_Library. The network of the Kellogg Library (@CSUSM_Library) of the California State University San Marcos was saved on July 20, 2014.

2. Identify clusters. Clusters within the networks were identified using the Clauset-Newman-Moore (CNM) clustering algorithm built into NodeXL. Fig. 2 can offer insight into the concept of clusters within a network. One might be tempted to partition the network into two groups, vertices A-D and F-I, connected by vertex E. These two groups could be considered clusters of vertices within the network, although placement of vertex E is not obvious. Application of the CNM algorithm identifies two clusters, A-D and E-I. In larger networks, visually partitioning the network into clusters is not usually possible and algorithms and computers are required.

3. Calculation of metrics. NodeXL was used to calculate the in- and out-degree, eigenvector centrality and betweenness centrality of each vertex in the network. General metrics describing the network as a whole were also calculated, including the average and median in/ out degree for the network as well as reciprocity, the likelihood that two connected vertices follow each other. These data were saved in the Excel file containing the network information.

4. Visualization of networks. NodeXL has powerful network visualization tools; however, we also used Gephi3, another network analysis and visualization software package to produce some of the network graphs. In these cases, network data was exported from NodeXL into a GraphML file. This file was imported into Gephi.

RESULTS

GENERAL NETWORK PROPERTIES

The follower/followee Twitter network for the library at The Richard Stockton College of New Jersey (RSC) is shown in Fig. 3. It has 205 vertices, i.e. 204 accounts follow or are followed by the library's Twitter

2 http://nodexl.codeplex.com/.

3 www.gephi.org.

account (or both). The network is constructed from 2485 edges, resulting in an average degree (both in and out) of 12.1. The median in-degree for the network is 3 while the median out-degree is 7. Reciprocity, i.e. the likelihood of finding two accounts that follow each other, was also measured for the RSC library network. A higher value indicates that there are more bidirectional edges, suggesting a strong relationship between the vertices (Hansen et al., 2010, p. 154). The library network has a reciprocity value of 28%. Finally, the CNM clustering algorithm identified four clusters, although one only contains two vertices. The first primary cluster includes the @RSC_Library account as well as accounts of professors, academic departments and other library related accounts. The other two primary clusters are dominated by accounts associated with the college, such as RSC's official Twitter account, and those associated with student organizations, such as the student senate. Personal student accounts are dispersed throughout the three primary clusters. See the Supplementary data for network graphs that segregate the clusters.

Interestingly, the network for the library at the California State University San Marcos (CSUSM) has values similar to those reported above for RSC's library (see Table 1). Five primary clusters were identified for the CSUSM network, however. An analysis of cluster membership did not identify specific populations inhabiting each cluster. Fig. 4 shows the CSUSM network.

METRICS

The values of in-degree, out-degree, eigenvector centrality and betweenness centrality were calculated for each vertex in the RSC and CSUSM networks. The vertices were then ranked according to the values of each metric. Tables 2 and 3 show the ten most important vertices in the RSC and CSUSM networks, respectively, as determined by each metric. The library accounts themselves are not included in the table. They are, by definition, the most important vertices in their networks; they connect to all accounts. The purpose of the study is to identify the other accounts which could be useful in disseminating library information. Out-degree is not included in the tables since it describes the number of accounts the vertex in question follows. Perhaps there are certain situations in which a large out-degree might imply importance, e.g. one might argue that a person who listens to many could be significant; however, out-degree matters little for situations concerning libraries and information dissemination. The values of the metrics are included next to the account names. The in-degree indicates the number of followers of that account. One is typically concerned with relative values of eigenvector and betweenness centralities. It should be noted that all of the accounts found in Tables 2 and 3 are associated with the institution; there are no personal accounts of faculty, staff, or students. Each account is maintained by a division, a department, or a program at the institution. These are the types of accounts which are influential in academic library Twitter networks.

Table 4 identifies the accounts whose values rank them in the top ten for each metric category. There are five in the RSC network and nine in the CSUSM network. Each primary cluster found in the networks contains a least one of these important accounts or the library account itself. That is, each cluster plays host to at least one important hub, which wields considerable influence over the cluster.

Table 1

Network wide metrics for the RSC and CSUSM library networks.

Metric RSC CSUSM

# of vertices 205 244

# of edges 2485 2257

Mean in/out degree 12.1 9.3

Median in-degree 3 1

Median out-degree 7 6

Reciprocity 28% 23%

# of groups 4 5

Fig. 4. The CSUSM network. Same features as Fig. 3.

DISCUSSION

Each of the three metrics used in this study indicate that institutional accounts, such as those maintained by the college/university, departments, and programs, are the most influential accounts in both the RSC and CSUSM follower/followee Twitter networks. The fact that this is true for the networks of both libraries suggests that it is a general feature of academic library Twitter networks, at least for medium sized, primarily undergraduate institutions like RSC and CSUSM. Further, these accounts are also likely to be influential in the wider institutional network, making them vehicles for information dissemination beyond the direct reaches of the library account. Recall that the metrics measure different definitions of importance. In-degree is the total number of

incoming connections to an account; it represents the amount of attention the account receives. Eigenvector centrality is based on the premise that connecting to influential accounts bestows more influence than connecting to unimportant accounts. Finally, accounts with high be-tweenness centrality act as bridges; they exist on many short paths between vertices, which can allow information to reach many with just a few retweets. Accounts with high values of each of these metrics are connected to many highly connected accounts and can allow information to pass efficiently through the network. This study demonstrates that institutional accounts in the RSC and CSUSM library networks have these properties.

Interestingly, the most influential accounts are not those of faculty, staff or students. Influential accounts, generally, are also not the ones

Table 2

RSC network accounts ranked in order of decreasing value of the metrics. The metric value is included in parentheses next to the account name. The median values are in parentheses next to their corresponding metric in the column title.

In-degree(3)

Eigenvector centrality (0.0031)

Betweenness centrality (2.4)

Stockton_edu (99) RSCCampusCenter (87) stocoargo (63) StocktonPAC (62) StocktonStuDev (60) StocktonDining (60) Stockton_GRAD (60) StkCollAlumni (60) SET_Stockton (58) stocktonospreys (56)

Stockton_edu (0.0200) stkbuzz (0.0165) litrscnj (0.0165) WGSSatRSCNJ (0.0165) RSCServiceLearn (0.0161) RSCCampusCenter (0.0159) peprsc (0.0158) Stockton_GRAD (0.0158) StkCollAlumni (0.0157) stocoargo (0.0154)

Stockton_edu (2831) stkbuzz (1222) RSCCampusCenter (1158) StkCollAlumni (660) litrscnj (652) stocoargo (480) Stockton_GRAD (467) ThePathYearbook (456) stocktonospreys (375) RSCGrantsOffice (361)

Table 3

CSUSM network accounts ranked in order of decreasing value of the metrics. The metric value is included in parentheses next to the account name. The median values are in parentheses next to their corresponding metric in the column title.

In-degree (1)

Eigenvector centrality (0.0027)

Betweenness centrality (0.8)

csusmnews (174) CSUSMCC (119) csusmHOPE (109) CSUSM_USU (91) CSUSMEL (85) ASLCSUSM (75) CSUSM_Greeks (67) CSUSMAdvising (63) CSUSMdiningserv (57) csusm_chabss (51)

csusmnews (0.0232) CSUSMCC (0.0191) csusmHOPE (0.0187) CSUSMEL (0.0166) CSUSM_USU (0.0163) ASLCSUSM (0.0157) CSUSMAdvising (0.0157) CSUSM_Greeks (0.0147) csusm_gradstudy (0.0140) CSUSMdiningserv (0.0131)

csusmnews (9691) CSUSMCC (3237) csusmHOPE (2601) CSUSM_USU (1351) CSUSMEL (1250) CSUSMAdvising (858) CSUSM_Greeks (604) ASLCSUSM (603) csusm_chabss (479) CSUSMdiningserv (465)

Table 4

Accounts ranked in the top ten for all three metrics.

RSC account name

Account description

Stockton_edu

RSCCampusCenter

stocoargo

Stockton_GRAD

StkCollAlumni

Primary institution Campus center Student newspaper Primary graduate programs Alumni

CSUSM account name

Account description

csusmnews

CSUSMCC

csusmHOPE

CSUSM_USU

CSUSMEL

ASLCSUSM

CSUSM_Greeks

CSUSMAdvising

CSUSMdiningserv

Primary institution Career center Student health center Student union

Academic community outreach Student organization Fraternity & sorority orgs. Undergrad. advising center Campus dining services

with the largest number of total followers, nor are they the most active on Twitter. The accounts with the most total number of followers are generally those that are known nationally or internationally, e.g. the Library of Congress and Hootsuite (a social media management platform), but are not necessarily central to the library networks. The most active Twitter accounts are those of individuals. There are two exceptions to this in the CSUSM network. The official account of the institution (@CSUSMnews) ranks highly in total number of followers, and @CSUMEL is the most active on Twitter (based on total number of tweets).

The metrics used in this study determine influential accounts based on the structure of the network, i.e. based on the wiring of the connections. Institutional accounts are wired to be important. Kim et al. (2012) discovered that the institutional accounts retweeted more library information than any other group; their behavior is significant. Combining these findings, one can conclude that such accounts are of paramount importance to the library networks. Their placement in the network sets them up to be influential (current findings). Then, their behavior allows them to capitalize on their position (Kim et al.). The results from the present study also highlight the importance of active participation by a library's account to the institutional Twitter conversation. As an institutional account itself, it is in a prime position to disseminate information from sister accounts within the college or university. This is also supported by the findings of Kim et al. Their study determined that, in addition to retweeting the most library information, the institutional accounts spawned more retweets (of their retweets) than any other group.

This information can be exploited by librarians for information dissemination purposes. Yep and Shulman (2014) show that partnering with highly connected accounts and agreeing to retweet messages can significantly increase exposure to library content. For example, if the library is hosting an event of interest to the campus as a whole, it might contact the primary account for the institution and request that its messages get retweeted. Both the primary accounts for RSC and CSUSM have many followers, most of which are affiliated with the institution and are potential recipients of the information. Alternatively, if the library obtains access to a new literature database, it could contact the account for the literature department. A simple click of the retweet button by the literature account will pass the message directly to those who are interested. Of course, it is not suggested that this be the only method of outreach. The library would probably do well to contact the literature department and faculty directly; however, using Twitter in this way requires minimal investment and will likely result in a more direct line to many literature students.

These examples, and the densely connected networks shown in Figs. 3 and 4, suggest that the reach of a library's Twitter program extends beyond its direct followers. Therefore, the number of followers is a poor estimation of the value or impact of a library's Twitter account. The environment in which it resides, i.e. the network, can play a significant role. Messages are capable of spreading broadly if retweeted by influential accounts, which are well connected in the library network and have many connections outside of it. The methodology presented above identifies these accounts and, notably, enables the quantification of the rather abstract concept of importance.

CONCLUSIONS

This work describes an analysis of the follower/followee Twitter network for academic libraries housed within two medium sized, primarily undergraduate institutions. Three network metrics, describing different measures of importance, were calculated for each account in the networks. This allowed for the influence of each account to be quantified. By all measures, institutional Twitter accounts were found to be the most influential in both networks, implying that this is a general feature of Twitter networks of academic libraries. The library Twitter network is a subset of a broader network encompassing the institution and its other divisions. The smaller library network was used to determine the significance of the institutional accounts; however, their influence extends beyond the confines of the library's Twitter neighborhood. Thus, the institutional accounts can be used to propagate information into the broader network of the institution. Partnering with the appropriate accounts can quickly direct information to a target audience.

This study shows that influence commanded by the institutional accounts is granted by the wiring of the connections in Twitter. Kim et al. (2012) found that such accounts retweet more library information than any other group. Thus, their behavior allows them to capitalize on their position in the network. Well connected, yet silent, accounts cannot propagate information, nor can vocal but poorly connected accounts. The results of Kim et al. combined with those of the present study demonstrate that institutional accounts are both well connected and vocal. The library, as an institutional account itself, can play an important role in information dissemination within the institutional Twitter network. This work, like the others that came before it, suggests that librarians managing Twitter profiles interact with other accounts on campus and take advantage of Twitter-specific features in order to contribute to the institutional Twitter conversation.

Librarians can adopt the simple procedure outlined above to map the networks of their home libraries, identify influential accounts, likely institutional, and rank them. Dickson and Holley (2010) note the time intensive nature of social media activities. It's true that managing a successful social media campaign requires a considerable investment in time; however the analysis described here does not contribute to it. Apart from the data download, which can take a few days while running in the background, the automated analysis with NodeXL can be completed in a matter of seconds with only a few clicks of the mouse. Such an analysis can allow librarians to tailor their Twitter outreach activities. In particular, it can complement a detailed examination of followers such as the one completed by Sewell (2013). One of the drawbacks of a microscopic examination of followers is the time requirements for manually obtaining follower information. NodeXL can help automate the import of much of the information, e.g. the numbers of followers and tweets, date of account creation, account description, among others.

ACKNOWLEDGEMENTS

This work is dedicated to the memory of Barry Alexander Moores. The authors would like to thank Talitha Matlin and Carmen Mitchell

from the Kellogg Library at CSUSM for assistance interpreting data from their institution. This work benefitted from discussions with Mary Ann Trail and Stephen Tsui. The authors also appreciate the Social Media Research Foundation for producing and providing access to NodeXL, the software package used in the study.

APPENDIX A. SUPPLEMENTARY DATA

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.acalib.2014.12.004.

REFERENCES

Cassidy, E., Britsch, J., Griffin, G., Manolovitz, T., Shen, L., & Turney, L. (2011). Higher education and emerging technologies: Student usage, preferences, and lessons for library services. Reference & User Services Quarterly, 50(4), 380-391. Cassidy, E.D., Colmenares, A., Jones, G., Manolovitz, T., Shen, L., & Vieira, S. (2014). Higher education and emerging technologies: Shifting trends in student usage. The Journal of Academic Librarianship, 40(2), 124-133. http://dx.doi.org/10.10167j.acalib.2014.02.003. Cole, S. (2009). 20 ways for librarians to use Twitter. Library Journal, 134(11), 25. Cuddy, C., Graham, J., & Morton-Owens, E.G. (2010). Implementing Twitter in a health sciences library. Medical Reference Services Quarterly, 29(4), 320-330. http://dx.doi.org/ 10.1080/02763869.2010.518915. Del Bosque, D., Leif, S.A., & Skarl, S. (2012). Libraries atwitter: Trends in academic library tweeting. Reference Services Review, 40(2), 199-213. http://dx.doi.org/10.1108/ 00907321211228246.

Dickson, A., & Holley, R.P. (2010). Social networking in academic libraries: The possibilities and the concerns. New Library World, 111(11/12), 468-479. http://dx.doi.org/ 10.1108/03074801011094840. Easley, D., & Kleinberg,J. (2010). Networks, crowds, and markets: Reasoning about a highly connected world. New York: Cambridge University Press.

Glazer, H. (2009). Clever outreach or costly diversion? An academic library evaluates its Facebook experience. College & Research Libraries News, 70(1), 11-19.

Gunton, L., & Davis, K. (2012). Beyond broadcasting: Customer service, community and information experience in the Twittersphere. Reference Services Review, 40(2), 224-227. http://dx.doi.org/10.1108/00907321211228282.

Hansen, D., Shneiderman, B., & Smith, M.A. (2010). Analyzing social media networks with NodeXL: Insights from a connectedworld (1st ed.). Burlington, MA: Morgan Kaufmann.

Kim, H.M., Abels, E.G., & Yang, C.C. (2012). Who disseminates academic library information on Twitter? Proceedings of the American Society for Information Science and Technology, 49(1), 1-4. http://dx.doi.org/10.1002/meet.14504901317.

Luo, L., Wang, Y., & Han, L. (2013). Marketing via social media: A case study. Library Hi Tech, 31(3), 455-466. http://dx.doi.org/10.1108/LHT-12-2012-0141.

Milstein, S. (2009). Twitter for libraries (and librarians). Computers in Libraries, 29(5), 17-18.

Newman, M. (2010). Networks: An introduction (1st ed.). Oxford; New York: Oxford University Press.

Park, J. -H. (2010). Differences among university students and faculties in social networking site perception and use: Implications for academic library services. The Electronic Library, 28(3), 417-431. http://dx.doi.org/10.1108/02640471011051990.

Petrovic, S., Osborne, M., & Lavrenko, V. (2011). RT to win! Predicting message propagation in Twitter. Fifth International AAAI Conference on Weblogs and Social Media (Retrieved from http://www.aaai.org/ocs/index.php/ICWSM/ICWSM11 /paper/viewPaper/2754).

Sewell, R.R. (2013). Who is following us? Data mining a library's Twitter followers. Library Hi Tech, 31(1), 160-170. http://dx.doi.org/10.1108/07378831311303994.

Suh, B., Hong, L., Pirolli, P., & Chi, E.H. (2010). Want to be retweeted? Large scale analytics on factors impacting retweet in Twitter network. 2010 IEEE Second International Conference on Social Computing (SocialCom) (pp. 177-184). http://dx.doi.org/10. 1109/SocialCom.2010.33.

Yang, J., & Counts, S. (2010). Predicting the speed, scale, and range of information diffusion in Twitter. Fourth International AAAI Conference on Weblogs and Social Media (Retrieved from http://www.aaai.org/ocs/index.php/ICWSM/ICWSM10/paper/ view/1468).

Yep, J., & Shulman, J. (2014). Analyzing the library's Twitter network: Using NodeXL to visualize impact. College & Research Libraries News, 75(4), 177-186.