Unraveling the Mysteries of Your Twitter Network

(This piece was originally published on LinkedIn on December 3, 2014)

1799ad7.png

  • How much do you know about your Twitter network?
  • How do you identify key influencers, find out what they’re interested in – and attract their attention?
  • How do you measure the interactions you can’t control?
  • Does follower count ‘really’ equal influence?

By Kirk Englehardt (@kirkenglehardt)

The magic of Twitter. It’s how I met Marc Smith (@marc_smith ), a sociologist specializing in the social organization of online communities and computer mediated interaction.

He grabbed my attention by sharing that I was a key influencer in the science communication twitter-verse, which came as a bit of a shock. I only joined Twitter two years ago…and I don’t even have 3,000 followers.

As I asked more questions, I learned much about how science communicators interact as a network on Twitter. I then I asked Marc to participate in this Q&A to help you uncover the mysteries of your own Twitter network. Schools, organizations, big brands and others will see value in this research. And it’s easy to do!

1. This summer you did a detailed analysis of the network of science communicators on Twitter. Why?

I am a sociologist interested in understanding social media. I view social media through the lens of several sociological traditions, particularly social network analysis (SNA). Networks form whenever things can connect to other things, and social networks form whenever people connect to people. Sociologists want to understand the size, shape, key features, and dynamics of social networks. A significant amount of research, methods, and tools have been created to reveal the nature of social networks. Social media networks are a particularly interesting sub-type of social network. Whenever people message or mention, follow or friend, rate, reply or review another person (among other things) they create connections. Collections of these connections form networks.

I work with the Social Media Research Foundation, a not-for-profit dedicated to creating tools and data that will enable a better understanding of social media and social networks.

The SMR Foundation produces the NodeXLapplication which extends the familiar Excel spreadsheet to enable the collection, analysis, visualization and reporting on networks. NodeXL
has built in collectors and data importers that connect to a growing number of social media network data sources, including Twitter, Facebook, YouTube, Flickr, email, WWW, wikis, blogs, and other platforms. With a few clicks, non-programmers can now easily extract data and automate its analysis to get a summary visualization and report that captures the shape, structure, key people and topics active in a network created from a collection of messages.

NodeXL is like a point-and-shoot digital camera for social media: it quickly takes a photo of the crowds that have formed in online spaces.

NodeXL makes it easy to access public social media data and generate a visualization and analysis; as a result, we make lots of social media network maps! We look for interesting topics and map them to see how different topics from different kinds of network structures.

2. The network diagram for #scicomm looks like a giant bowl of green spaghetti. What, exactly, am I looking at here?

(View a larger version of this graphic with supporting data)

A social media network is composed of people, or user accounts, which may or may not connect to one another. In a NodeXL network visualization, each person in the data set is represented by their profile photo, with a size proportionate to their follower count. The green lines are the connections among people in the network. Social media enables people to connect with one another in many ways. In Twitter, people connect to one another when they tweet each other’s names (@username). Some people get mentioned or replied to by large numbers of other people but most people attract few to no connections. The emergent pattern of collections of connections forms a shape that tells us a great deal about the social behavior within the population. Many networks are composed of sub-groups or clusters that can be identified using a variety of algorithms.

If there are two large dense clusters with little interconnection, the network may be a polarized or divided topic. If there is just one dense group, the network may be a unified in-group community. A network in which a significant portion of the population have no connections at all (they never mention other people by name) is fragmented and is often associated with brands and public topics. Some fragmented brand topics gain some clusters of connected people forming a community clusters pattern. When people mostly repeat (or retweet) what one or a few accounts post, a hub and spoke pattern is created that reflects a broadcast of information. The many people who retweet the hub account rarely if ever connect to one another, forming an audience rather than a community. The mirror opposite, the out-hub-and-spoke pattern, is created when a single account replies to many accounts which, again, do not connect to one another. The difference is the direction of information flow, in an out hub-and-spoke pattern network the hub is performing a support function by answering many people’s questions. Within these different network structures, people occupy various positions, some of which may be strategic or rare, like the few people who are in the middle or center of these networks. There are several measures of centrality in network theory, and we find that a measure of how much a person plays the role of a “bridge”, known as “betweenness centrality”, is a useful way to highlight people who play an influential role in the network.

Looking at the SciComm NodeXL social media network map and report (see the details here) we can now see that the network is largely dominated by the broadcast hub and spoke pattern. This network is composed of the connections among the people who tweeted “#scicomm” over the 34-day, 21-hour, 40-minute period from Tuesday, 23 September 2014 at 03:03 UTC to Tuesday, 28 October 2014 at 00:44 UTC.

This network pattern suggests that #SciComm is not a “brand” since just 169 out of 5,803 Twitter users whose tweets in the requested range contained “scicomm”, were isolates – group 9 contains this fragmented portion of the population. A brand network would have a larger fraction of the population be isolates, often the largest cluster in the network is populated by disconnected users in a brand network.

#SciComm is, therefore, a broadcast network in which ten prominent accounts which have attracted the most diverse retweet audiences. The top ten vertices, ranked by betweenness centrality (influencers) in this network are:

  1. @kirkenglehardt
  2. @shiplives
  3. @dnlee5
  4. @thilinah
  5. @fromthelabbench
  6. @bisgovuk
  7. @lizneeley
  8. @rjmlaird
  9. @whysharksmatter
  10. @trevorabranch

These very central people are likely to be influential in the network – they attract connections most widely across the network. We like to think of them as they “mayors” of a hashtag – the informal leaders who drive the conversation by nominating topics that get picked up and repeated by the widest group of people.
Content used in the network can also be analyzed and ranked by frequency of mention.

Ranking the top ten hashtags mentioned in tweets in the #SciComm network were:

  1. #scicomm
  2. #science
  3. #cdnsci
  4. #sciwri14
  5. #socialpeer
  6. #scifund
  7. #cdnpoli
  8. #sciox
  9. #sciart
  10. #ebola

These are the topics most closely related to #scicomm. Similarly, we can look at the most frequently mentioned URLs and discover what websites and blogs were of greatest interest in the #scicomm discussion network:

  1. http://www.theguardian.com/science/blog/2014/oct/10/science-communicators-quantum-physics-granny
  2. http://elifesciences.org/content/3/e04902
  3. http://newsoffice.mit.edu/2014/importance-of-science-writing-0923
  4. http://blogs.scientificamerican.com/urban-scientist/2014/10/27/if-you-cant-be-a-good-example-be-a-warning-how-ecointernets-scicomm-fail-can-make-you-a-more-culturally-aware-science-communicator/
  5. http://paper.li/rjmlaird/1383841564
  6. http://www.americangeosciences.org/policy/internships-and-fellowships#Schlum
  7. http://www.scilogs.com/communication_breakdown/outreach-and-impact-2014/
  8. http://www.scicomm.it/p/pbt-2014.html
  9. https://www.linkedin.com/pulse/article/20141010181247-3091133-the-week-s-top-science-communication-stories-10-10-14?published=t
  10. https://evidencefordemocracy.ca/sciencewatchdog

In the detailed report that is generated by NodeXL along with the network visualization, we duplicate this data for each sub group in the network. This is a useful way to get a sense of the overall topics discussed, identify the key people, and recognize the ways different groups focus on different topics. In a world of too many tweets, a network map and content summary report may be a better way to maintain an overview of a collection of Twitter users.

3. What did the analysis tell you about science communicators who tweet? 

The #SciComm Twitter network map and report shows that this is primarily a broadcast community that has modest visibility beyond its existing “usual suspects”. Not many people casually mention #SciComm and most of the time people who do mention the hashtag do so by retweeting someone else’s use of the term. Sharing is happening, but not necessarily “community” or conversation. In a network, a community is a more dense web of interconnections, with many cross-connecting links rather than a hub and spoke audience structure.

4. In September, Science Magazine published a list of the Top 50 Science Stars of Twitter. It created an uproar because it ranked people by follower count…but it seems follower count has little to do with influence.

I agree. The count of followers is not a good proxy for influence. One issue is the global/local tension. Followers could be attracted to an account for one topic and offer no insight into the value of a user on a different topic. So we want a more “local” measure of influence. We want to know how influential a person is on a specific topic, not just globally.

It is more useful to calculate a measure of influence within a specific network. We can create data sets that have meaningful boundaries; for example a collection of messages and people who talk about a specific topic, like #SciComm.

Within a meaningfully bounded network, we might also want to recognize that there are different kinds of influence or at least different kinds of valuable positions within the network. Many people want to identify “influentials” – they key people who drive the conversation. Network analysis makes it easy to find these people – they are at the center of the network. But networks can also identify people at the periphery who may also have value, perhaps for bringing in new members or finding new customers. For that goal, the people at the center are of limited value: they already know about you.

So a more sophisticated notion of influence might consider a bounded data set and look for people who occupy different kinds of positions within it. The list of mayors we identified for #SciComm are clearly interesting: these are the leading voices in the #SciComm network. But network analysis suggests that people in positions like isolates, the people in Group 9 in the #SciComm network, who have just casually mentioned #SciComm and have no connections yet, are also important.

5. When we spoke you called me ‘The Mayor of the #Scicomm Hashtag” – but I don’t have that many followers and I don’t tweet as much as many of the others in the network. What does it mean?

You are the “mayor” of #SciComm: you rank #1 in terms of your “Betweenness Centrality” – you are most in the middle of the network. We did not count followers in calculating this metric – it was irrelevant to answering the question “How do other people connect to you in this topic network?”.

Someone with very few followers who says something that gets other’s attention can be the mayor of a hashtag. Network analysis provides a way of ranking social media participants by behaviors they do not themselves control.

People can tweet more or less, but it is hard for them to control how many other people reply or retweet them. Network analysis looks at how others act towards you rather than on how you act yourself. Like the way Google calculates PageRank for all web pages, network metrics are somewhat resistant to manipulation because they require extensive collusion to distort. That imposes a cost that most people do not want to pay to overcome (though spammers and others are happy to do so.)

6. While the diagram shows one community, science communicators, it looks like it’s broken into sub-groups, what can we learn from that?

Most networks are composed of neighborhoods or sub-communities. In some cases, when these sub-groups have very little interconnection, the divisions between groups can be an indication of conflict or polarization. But in the #SciComm network, these sub-groups are interconnected neighborhoods that form around the leading #SciComm “mayors”. Each mayor has developed an overlapping but distinct audience. People engage in #SciComm by mostly retweeting these key people. Each group has a slightly different focus which is visible in the various hashtags and words used in each group. Some groups are talking about a recent conference (SciWri14) while others are discussing physics or funding opportunities. The #SciComm discussion is a diverse one and this is reflected in the ways each sub-group uses different hashtags and words.

7. How can a communications person use this kind of analysis? What does it mean for schools, organizations, companies and brands? 

Maps are useful things. You can select a destination with a map as well as plot a course to get there. Social media network maps are ways to illuminate the landscape of social connections. When seen as a network map social media becomes less confusing.

While network visualizations are often complex images, with some effort people can develop a visual literacy needed to read these images and get value from them. Like many scientific images, they reflect an underlying complexity as simply as possible.

Communications professionals can use social media network maps and tools like NodeXL to more carefully focus their efforts. First, what is the nature of your existing network? What about the networks of related topics, issues, competitors, and events? Who are the key people in these discussions? NodeXL identifies the mayors of your hashtag so that you can quickly follow these influential voices. Part of the NodeXL report is a summary of the words and hashtags that most reflect the interests of each person. Using these terms when tweeting messages at the mayors is a good way to get their attention and get them to hear your message.

The goal is to have the most central people in the conversations most relevant to you to repeat your messages, amplifying their reach and visibility by orders of magnitude.

(Details of this graphic)

At the same time, many communication strategies also want to focus on the other end of the network influence spectrum, by engaging the isolates to get people with a casual interest to return and increase their involvement with the topic. Most social media marketing analysis services seem to ignore these peripheral people to focus exclusively on the “mayors”.

Communications professionals may want to recognize the overall shape of their topic networks and ask if the shape is optimal for their goals. Creating a brand on social media, in which many isolated people mention a product or service, is a meaningful goal. But weaving those disconnected people into a community that begins a conversation may be a more advanced goal. Shifting the shape of the network is a way to leverage the network maps over time. Various interventions, like hiring new people, buying ads, or using a new service, can be evaluated by making network maps before, during and after and looking for structural changes in the network.

Brands and companies can quickly map a range of relevant topics every day using NodeXL, even receiving updated maps and reports on your phone on a scheduled basis. This becomes a powerful way to consume the traffic generated by 10-20 fast moving topics without being consumed by the need to read 100K tweets per day. Social media network maps and reports quickly summarize the shape of the crowd, its key leaders, and the topics that they care about into a quickly digestible form. Using these maps to navigate social media, communications professionals can more quickly select destinations and chart a clear course to engage them.


Additional Resources:

Additional Research: 

Connect:

(This piece was originally published on LinkedIn on December 3, 2014)

One thought on “Unraveling the Mysteries of Your Twitter Network

  1. Pingback: Scraping Twitter follow networks: or how to win at social media – RealThinks

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s