Museums on social media: Analyzing growth through case studies

Alex Espinós, La Magnética, Spain

Abstract

Keywords: Social Media, Analytics, Social Network Analysis, Facebook, Twitter, SNA

1. Introduction

During the Baltimore and Florence conferences, we presented our report on how two thousand museums worldwide related on Twitter. We analyzed the most influential museums on Twitter, the factors underlying the observed community structure, the evolution of museums’ Twitter use during the study, and the museums that worked as a hub connecting communities from different countries.

Last year, as part of a broader research, we focused on three case studies about the relationship between a museum and its followers and how to use this information to design better social media strategies.

In this paper, we focus the last part of this three-year-long research about museums on social media on the growth of a museum’s community in social media, including causal factors and quantitative and qualitative issues.

We will close the paper with a review of how museums’ follower bases have evolved in a two-year lapse: from December 2013 to December 2015. We will analyze some factors explaining the observed results and the implications they have for museums, particularly for smaller and medium ones.

2. New followers’ acquisition mechanisms

A museum can acquire a new follower in several ways on Twitter:

• Recommendations made to users who just signed up. A new user signs up on Twitter and is asked about the topics he or she is interested in. After setting a location and choosing museums or culture among their topics of interest, a list of museums’ profiles are shown so the new user can choose which ones he or she wants to follow. After choosing several museums, other museums are going to show as suggested Twitter accounts to follow. This second list depends on the accounts first chosen and is a particular case of the next mechanism.
• Other recommendations. New profiles to follow based on our past behavior—who do we follow, mention, favorite, and retweet—are continuously offered. Email recommendations are included in this point.
• Mentions or retweets. When one of our followers mentions us in a tweet or retweets us, this follower’s followers see us and may choose to follow us. Tweets that begin with the mention are an exception, as they are not shown on our followers’ timelines.
• People following a hashtag such as #museumweek or #askacurator see our tweets and decides to follow us.
• People with previous knowledge of the museum or who have visited its website go to Twitter and follow us.
• Twitter ads.

Most of the paper is devoted to showing that the first three mechanisms (recommendations made to users just signed up, other recommendations, mentions and retweets):

• Are based on the same network dynamics feature, the triadic closure or, for short, triangulation. We will analyze this simple yet powerful mechanism and its deep implications to social media strategy. We are going to explain it in the next section.
• Account for most new followers. The exact percentage varies but is usually in the range from 70 percent to 90 percent. So, this is the source of most of the social media growth and information spread for museums, and this fact is a first finding the paper offers.

3. Triadic closure: A simple mechanism in a complex environment

One of the main social media growth mechanisms is called triadic closure. We will show empirical evidence that growth mechanisms related to triadic closure provide most new followers.

The underlying idea is quite simple: if two people are connected through a third person, chances are they can get connected in the future. It is the way we have gotten to know many of our acquaintances and friends in the “real world.” The chances that these two people or two Twitter accounts connect may still be small, yet they are much higher than those of two random users among the half a billion Twitter users or 1.4 billion Facebook users.

In fact, triadic closure is a broader concept, but we are going to focus mainly on a key aspect for museums in social media: the transitivity of the follow relationship: if A follows B, and B follows a museum, M, it is likely A will follow M. Relationships on Twitter are asymmetrical, so there are many different configurations of a triad of connected users. The schema shown in the diagram bellow is the minimum schema that allows triadic closure. The schema works the same if symmetrical relationships exist between A and B or between C and D.

Figure 1: the arrows show the direction of the relationship, as in Twitter relationships (follow, retweet, mention, favorite), is not symmetrical. Relationships between a Facebook page and followers are not symmetrical either.

Triadic closure is also related to other key social media aspects:

• Information spread. Information flows along the same connections but in opposite direction to the arrows depicting the follow relationship. For applying triadic closure to the description of information flow mechanisms, we would need to slightly rephrase it, but the underlying structures and ideas remain the same.
• Trust relationships also show some degree of transitivity following the same schema.

One may guess that the more common friends, the more likely two users are to connect, or at least that this probability increases until a certain saturation point. We will show some empirical results on this topic. The more common acquaintances we have with someone, the higher their profile is going to show in our LinkedIn list of recommended profiles. LinkedIn knows these profiles have an increased probability to be picked; otherwise, they will not keep showing them in the top of the page when they have machine-learning systems to show the options the users are more likely to choose.

Although triadic closure is a simple and easy to understand mechanism, it works over time on a very complex environment. The image below shows Italian museum Palazzo Madama connections on Twitter, including only the profiles that have either mentioned or retweeted the museum, or that have been mentioned or retweeted by it. It shows also the crossed connections between the Twitter profiles that have been connected to the museum. This graph is quite smaller to that depicting the follow relationship, as a museum only talks to a fraction of its followers. And this is an example of a museum that had around 8,000 followers at the time the analysis was performed for MW2015. The environment gets much more complex as the number of followers grows, as we showed in last’s year paper with Barcelona’s CCCB, which had around 100,000 followers among its eight Twitter accounts, and London’s Victoria and Albert, which at that time had 525,000 followers.

Figure 2: Palazzo Madama’s environment on Twitter.

4. Recommendation as a particular case of triadic closure

As Twitter (n.d., a, b) explains, tailored recommendations are based on two sources of information:

1. Users who are frequently followed by users who follow the accounts you follow. This is quite simple: many users follow all or most of London’s major museums; if I follow one or two of them, I am going to see other London museums as a suggestion. The same would apply to a topic-centered set of museums. As shown in the chart below, this is just another case of triadic closure. This same basic procedure is also used by Facebook or LinkedIn, although the fine tuning of the algorithm may differ.
2. The other source is based on the websites you have visited during the last ten days. Certain restrictions apply to this mechanism depending on your browser settings and on the number of devices—including smartphones and tablets—you use to access the Internet. The user can turn off this source of recommendation on Twitter settings. According to Twitter, this second source of information is not used in Europe, but we have found evidence pointing otherwise.

Figure 3: M1, M2, and M3 are three museums that share a significant number of followers. If a user follows M1 and M2, Twitter is going to recommend M3 to this user, because Twitter knows it is likely the user would be interested in M3. If either M1 or M2 are important museums with a large set of followers, this can be a significant source of new followers for M3. They key question relating to social media strategy is: how do we get our museum to get recommended to M1 or M2’s followers?

Besides these two sources of suggested profiles to follow, recommendations for new users also have another source: the data you provided when signing in (e.g., locations, topics of interest).

It is not possible to mimic the Twitter recommendation algorithm for museums without a large network of computers and special access to their API. But we can study the part of the recommendations that are traceable to a triadic relationship. Following the approach chosen in the former two papers of this research, we will focus on a stronger relationship than who follows whom; we will focus on a subset of followers, those with whom we have an actual relationship through mentions or retweets. This may seem an important restriction to the research, but results show otherwise, as we shall see shortly.

This subset of relationships and the triadic closure mechanism account for the vast majority of new followers in all museums analyzed. This is the second key finding of the paper: the actual influence, the spread of information, and growth chances are provided by a subset of your followers: those who really engage with you. This may be obvious regarding the new followers that discover a museum through a retweet by one of its followers, but it is not at all obvious when analyzing recommendations and other sources of new followers.

This result should turn some strategies from focusing on the total number of followers to strategies focusing on engagement and, as we shall see shortly, on followers’ diversity.

5. Mentions and retweets

Every follower of a museum has its own followers. This creates a larger set of people we cannot reach straightforward with our posts, but we can reach some of them through mentions and retweets from our followers.

Some notation may help:

• Let Followers(M) be the set of followers of a museum or, for short, F(M). It includes the museum, M. The blue circumference in the diagram below represents the followers, F(M).
• Let Followers(Followers(M)—for short F(F(M))—be the set of people following at least one of the museum followers. It includes the museum followers: F(M) is included in F(F(M)). This set matches the whole diagram: the museum at the center and the two concentric circumferences representing the museum’s followers (blue) and their followers (green).
• F(F(M)) – F(M) are the users that can be reached through mentions and retweets, but are not followers of the museum. This set is represented in the diagram below by the outer circumference in green. For each user in this set, we have at least a path, as shown in the following diagram:

Figure 4: blue dots represent museum followers, and dots in green some of their followers. The dashed arrows represent connections that may be created through triadic closure. There is a two-step path from the museum to FF1, but there are two such paths leading to FF2. Is there a higher probability FF2 will follow the museum? This is a question we shall try to answer.

The museum is going to get most of its new followers within set of users that are connected to the museum through a path of length 2. Some of them will get to follow the museum through a retweet or a mention from a user they follow that follows the museum; others through tailored recommendations picked by Twitter within this set.

We are not going to add further complexity to the notation, but we are focusing on a subset of F(M) and F(F(M)). As F(M) we are not chosing the full set of followers, but only those that have also mentioned or retweeted the museum. The diagram remains the same, but with the inner circumference including only the engaged followers.

Each time we gain a new follower, f, the set F(F(M)) increases. The increase the new followers provide depends on two factors:

• The number of his or her followers, F(f).
• The overlap that the F(f) has with the set of our followers—F(M)—and our followers’ followers—F(F(M)). There is usually a significant degree of overlapping; the less diversity our community has, the higher the overlapping.

These first two factors affect the number of potential new users we can reach through the acquisition of this new follower, but say nothing about how likely is that this new follower will actually increase the spread of our posts or help us to get new followers among those who follow him or her. For that, we need a measure accounting the activity and overall influence of a user. We can approximate it by its Klout, and it is a key factor to assess a user’s value in growing our community.

6. The main sources of new followers

Let us show a sample of the results of our empirical research:

Table 1: weekly new followers and their sources

The following chart shows the share that the three sources related to triadic closure have on the total museum followers’ growth:

Figure 5: share of followers’ growth due to triadic closure.

The results on the chart show the importance of mentions and retweets, and also the followers acquired through recommendations to new users. Nonetheless, the chart may be misleading because the differences in growth speed of the museums shown. The same chart with the weekly new followers acquired by each mechanism helps to better understand the weight of each source:

Figure 6: weekly new followers acquired through triadic closure.

The information in the table allows for some clear remarks that have direct implications in social media strategy:

• Most new users can be attributed to triadic closure. The exact percentage ranges from 68 percent to 94 percent in the cases studied. The restriction applied to the analysis focusing only on actual relationships—mentions, RTs—has helped to narrow the causes of growth.
• When museums gain new followers at a path of five hundred to several thousand new followers a week, the main source of new followers are Twitter recommendations. Being among the choice of tailored recommendations for new users from a certain location and within a topic of interest allows for a much faster growth. It draws a line between museums that receive most of their new followers from this source and grow faster, and those that do not. Tailored recommendations to other users plays a similar role, but its figures are usually lower.
• Surprisingly, mentions and retweets allow only for a slow growth and account for a small share of growth in fast-growing museums. And it is interesting to note that differences in new followers acquired though this source are much smaller than differences in overall followers’ growth. This is a third finding of this paper, an unexpected one.

The data collection and analysis procedures for these calculations are very complex, so we have limited to museums up to one hundred thousand followers.

7. Is there an increased probability to acquire a user as a follower when there are several paths to reach that user?

We can rephrase the question this way: the more common acquaintances, the more likely two users will connect? And if this is true, is there a point beyond which any new common acquaintance does not increase the probability of connection, or even decreases it? The idea behind the second question is: if two users have a lot of common connections, it is very likely they have had several opportunities to follow each other, but if they have chosen not to we can assume they have no interest in following each other. This is also related to the stagnation problem in museums with lack of diversity among their followers.

Figure 7: probability to acquire a user as a follower within a week depending on the number of triangles connecting the museum with the user.

The chart is rather technical, and we are going to focus on what can we learn from it:

• There is a clear increase in the probability to get a user as a new follower when there are several paths connecting the user to the museum (like FF2 in Figure 4).
• When there are three to five paths connecting us to the user through active museum followers (remember, we are only considering the followers that engage with the museum), the probability is quite high. A weekly chance of 0.6 percent to 1.9 percent is equal to a chance of 27 percent to 63 percent within a year.
• From six paths onward, our calculations are based on a much smaller sample, as it is not usual to have such a number of connections. The sample of users connected to a museum by six different paths has around seven hundred and fifty users, and there are ninety-five connected through nine paths, the last value we have chosen to show on the chart.
• There is some evidence to believe that after a certain threshold of about six different paths, the probability decreases. The evidence is nonetheless inconclusive. The stagnation observed in a large number of social media accounts provides indirect evidence pointing to the same fact.

8. Implications for a museum’s social media strategy

In this section, we are going to focus on some implications of these findings for a museum’s social media strategy.

Engage with the big players

A connection—reciprocal following, reciprocal mentions, or retweets—with a big player will create thousands of new triadic relationships that for the smaller museum are thousands of opportunities for growth and word spread. The big players that are more useful for a small or medium-sized museum are those nearby and those centered on the same topic.

This strategy is not feasible for private companies, but is possible in a collaborative environment. When relating with museums that have hundreds of thousands of followers, it is important to be aware that they usually behave—and need to behave—as a celebrity. It is impossible for them to manage all the mentions they receive or to engage in all conversations about them. You need to strengthen the bound slowly, and for that it is useful to use not only Twitter but also professional conferences, a more personalized social media connection through LinkedIn, etc.

Figure 8: two ways of representing the new opportunities that a small museum M1 gains when engaging with a large museum M2. The first diagram shows the potential new connections user by user, while the second shows them in a simplified and more abstract way.

Diversity to increase information spread and avoid stagnation

Diversity within the set of our followers—F(M)—and the set of our followers’ followers—F(F(M))—is important to ensure both growth and information spread. The same number of followers split in different geographic and topic-centered communities will lead to a greater growth and word spread than the same number of followers in a single local community.

Stagnation in follower growth is a usual problem. Many museums’ follower counts over time have this shape:

Figure 9: followers’ growth over time.

The museum starts slowly, and then its growth path speeds up, but after a threshold the growth is increasingly slower. Sometimes the stagnation is complete and the museum gains only a few followers a month but also loses a few; sometimes it leads to very slow growth taxes. The initial point may vary, and if a popular museum starts its presence on a new social network, it may have a bump of followers during the first few days or weeks due to other sources—website, press, their community on other social networks, such as part of our Facebook community following us also on Instagram—or word of mouth.

Stagnation is most often related to lack of diversity. The mathematical models to prove that are too complex for a paper that is aimed for a nontechnical audience and wants to focus on the implications of its findings for Social Media strategy.

Let us show the idea with a series of images:

Figure 10: from step 2 to step 6, we can grow steadily, but after a certain threshold our growth tax is going to quickly decrease due to the lack of new potential followers. We are going to exhaust this community: all users interested in our museum will follow us as, and after a while, we will have reached all users in the community several times; and those not interested (the blue dots in the last step) will not follow us. But we need to get new followers out of this community to ensure mid-term growth.

Museums that have communities in several countries and around different topics do not usually have this problem and may grow at a healthy rate of 40 percent to 80 percent a year.

Language diversity for museums outside of English-speaking countries is also important, as it allows you to engage with people interested in your museum worldwide and keep in touch with tourists who have visited your museum. Palazzo Madama’s results, based on their bilingual communication and how they compared to other Italian museums that posted only in Italian, or Barcelona’s CCCB trilingual posting are two clear cases analyzed in last year’s paper. Even for museums in the United States, it may sometimes be wise to use other languages in areas with significant linguistic minorities.

Use Twitter ads wisely: A small budget can make a great difference

The cost per new follower of a Twitter campaign depends on the country, month of the year, and target. For a well-targeted and designed campaign, they may be in the range of $0.30 to $1.00. If the campaign is not properly designed, the cost may increase up to tenfold or even more.

Most museums lack the budget to base their growth on campaigns, but many have small budgets than can be used wisely as a seed for allowing future growth. If campaigns are used to get to key influential users in new communities, our potential growth through the triangulation process described in this paper can increase greatly even with a small investment. If we think about the resources devoted to social media and the size of our audience, it is an idea worth considering.

Twitter campaigns allow different forms of advanced segmentation, including targeting a campaign to another museum’s followers. Facebook also allow advanced targeting, but has not this possibility. Both have re-marketing campaigns, which help to enroll our website’s visitors on social networks. As this kind of campaign targets only visitors to our website, they are inexpensive and effective for small and medium-sized museums.

The importance of the profile information

An appealing avatar and a clear and short name increase the probability of being chosen when appearing in a tailored recommendation piece such as this one:

Figure 11: Guggenheim’s avatar is immediately recognized. MoMA’s logo is visually strong, but National Gallery’s is almost invisible on this piece suggesting museums to follow on Twitter.

The museums more frequently picked in these lists are more likely to keep appearing as tailored recommendations, while those with lower rates will appear less often.

9. Museums on Twitter: Changes from December 2013 to December 2015

We have been monitoring museums on Twitter for over two years. For this survey of the evolution of museums on Twitter during the last two years (December 2013 to December 2015), we have filtered the initial set of museums, picking only those with significant Twitter activity.

We have chosen museums that:

• Had more than one thousand followers in both December 2013 and December 2015
• Have published at least one hundred tweets during this time scope; that is, an average of at least a weekly tweet during the last two years
• Have received at least fifty mentions or retweets during the time scope

There are 906 museums from the initial list that match this criteria.

Table 2: museums on Twitter: 2013 to 2015 comparison

The variance of the data has strongly increased in 2015, which means that differences among museums have grown. This may have many causes, but the tailored recommendation mechanism accounts for a significant share of this increase in differences.

When trying to relate the number of followers in 2013 with those in 2015, only a chart with both axes in logarithmic scales makes sense of the data. We mention this fact because it has implications on the factors underlying followers’ growth, but we are not going to dig into the technical details of this.

The fact that the increase is proportional to the number of followers has deep consequences in the increasing differences among museums with different sizes and budgets.

Average values give us an overall understanding of the phenomenon we are analyzing but hide the diversity of situations, which is better shown by the distribution below that shows the average increase (%) in number of followers depending on the number of followers museums had in late 2013:

Table 3: museums on Twitter: growth rates from 2013 to 2015 classified by the number of followers on December 2013

Roughly speaking, there are three groups of museums and growth rates:

• Those over 250,000 that have increased their followers by 80 percent within two years (average).
• Those ranging from 12,500 to 250,000. This is the group with a higher growth rate: an average of 131 percent.
• Those under 12,500 with slower rates: 84 percent.

When looking at absolute growth rates, the results are clear: the higher the initial number of followers, the higher the increase within this two-year time span. Differences between museums have grown.

One of the causes of the greater growth rate of the second group is the growth of some world-class French museums that ranked low on their social media activity on 2013. Our Museums and the Web 2014 paper focused on museums during a twelve-month span (April 2013 to April 2014), and one remark we noted about country communities was the low results achieved by French and Italian museums when compared to their rich heritage and large figures of on-site visitors. The situation has evolved since then, and some French museums are getting to rank according to their importance.

The table bellow shows the ten museums with a greater relative increase. Nine of them are French:

Table 4: top ten museums by their followers’ increase rate

The table below shows the top twenty-five museums by their absolute followers’ increase:

Table 5: top twenty-five museums by their absolute followers’ increase

Figure 12: followers increase in 2014-2015 for the top twenty-five museums worldwide.

There are significant differences from 2013 followers’ rankings:

• The United Kingdom and United States still have an edge over other countries, but differences are smaller.
• Early adopters such as the Netherlands and Sweden have lost their place in the rankings as larger museums from other countries have developed their social media strategies. In a ranking relating their social media results with their on-site visitors, they would still rank high.
• Large museums from Spain have found their place in the ranks.
• A latecomer such as France has several museums ranking high, but Italy still does not.
• Despite the high use of Twitter in Indonesia, Japan, Brazil, and other Latin American countries, the United States and Western Europe still dominate the rankings with few exceptions.

Acknowledgements

Gloria Castellví plays an important role on our Social Network Analysis research, has provided some of the data, and has reviewed the paper; Isaac Filella was responsible for gathering the data about triadic closure; Veronika Pasishna drew the diagrams showing triadic closure; Pietari Posti has helped me with correcting the text; and Silvia Trillo has reviewed the paper and has made useful suggestions to improve it.

References

Barabasi, Albert-laszlo. (2013–2014). Network Science. Ongoing project, Center for Complex Network Research Northern University. The printed version of the book will be published by Cambridge University Press in May 2016. PDF available at http://barabasi.com/book/network-science

Barabasi, Albert-laszlo. (2014). Linked: How Everything Is Connected to Everything Else and What It Means for Business, Science, and Everyday Life. Basic Books.

Easley, David, & Jon Kleinberg. (2010). Networks, Crowds, and Markets: Reasoning About a Highly Connected World. Cambridge University Press.

Espinós, Alex. (2014). “Do Museums Worldwide form a true community on Twitter? Some insights on the Museum Twitter ecosystem through Social Network Analysis and Network Science.” Museums and the Web 2014.

Espinós, Alex. (2015). “Museums on Twitter: Three case studies of the relationship between a museum and its environment. Victoria & Albert, Palazzo Madama, CCCB.” Museums and the Web 2015.

Golbeck, Jennifer. (2013). Analyzing the Social Web. Morgan Kaufmann/Elsevier.

Granovetter, Mark. (1973). “The Strength of Weak Ties.” The American Journal of Sociology 78(6): 1360–1380.

Granovetter, Mark. (1983). “The strength of weak ties: A Network Theory revisited.” Sociological Theory 1: 201–233.

Gulbahce, N., & S. Lehmann. (2008). “The art of community detection.” BioEssays 30(10): 934–938.

Jeong, H., Z. Neda, & A.L. Barabasi. (2003). “Measuring preferential attachment in evolving networks.” Europhysics Letter 61(4): 567–572.

Newman, Mark. (2010). Networks: An introduction. Oxford University Press.

Onnela, J.-P., J. Saramäki, J. Hyvönen, G. Szabó, M A. de Menezes, K. Kaski, A.-L. Barabási, & J. Kertész. (2007). “Analysis of a large-scale weighted network of one-to-one human communication.” New Journal of Physics 9: 1–27.

Twitter. (n.d., a). “FAQs about tailored suggestions.” Available https://support.twitter.com/articles/20169941?lang=en

Twitter. (n.d., b). “About Twitter’s suggestions for who to follow.” Available https://support.twitter.com/articles/227220