Conference Twitter Mapping to Support Communications Strategies
This is a continuation of a discussion of mapping Twitter networks generated at conferences and events as a way to support communications strategies and to determine the effectiveness of the event at both engaging and bridging audiences. We looked at the social networks outlined by Twitter messages from the IMPCON 2016 and American Evaluation Association Annual Conference 2016, held last October in Atlanta. Today we will look at how the findings of Twitter mapping can be used for communications strategies.
Let’s look at communities first. Communities generally, and more specifically, clusters, as in the
first figure below, are collections of nodes that are relatively more interconnected with each other than they are with the larger network. This generally indicates some kind of commonality of interest or affiliation. Often it is possible to determine what this underlying factor is, either heuristically, or if the data is available, using tests developed in the network analysis literature.
For example, if we want to understand the overlap with social investment, we can look more into what the G1 and G2 (the two blue clusters) members said in their communications, and if we have the data, their affiliations, sectors, geographic locations and so on.
In the future, AEA might want to look at their members’ positions relative to other events or issues, their tendency to cluster and the factors that may explain these clusters.
The existence of communities, or clusters, in a network can have implications for the diffusion of messages and information.
If members of a cluster are very similar in some way (technically called “homophily”), they may either be relatively receptive or unreceptive to a message. If unreceptive, then for example, they may impede the spread of information by not being interested in passing it on. This will be particularly important if a cluster is located on the paths between other members of the network. Knowing this can help shape and target messaging in order to maximise diffusion. Knowing a community’s interest is also valuable for shaping messages that are meant specially for them.
Remember from the last post that we’ve also identified the central group, many of whom are in cluster G1 (dark blue), as important communicators. It helps to identify these highly connected players in order to target communications and increase diffusion efficiently. Selecting the first-degree connections of the central group in cluster G1 shows that they are attached to a large part of the wider network. The central group in G2 (light blue) is important too, but not quite as much as the G1 group.
Metrics: Degree and Centrality
The members of a network are characterized by a number of metrics describing their relationship with other members of the network and their status relative to the network. The nodes in this mapping are sized according to their in-degree. We’ll discuss this and several other key vertex (node) metrics.
The following chart presents several key metrics that we can use in identifying members of the network who should be particularly important to information diffusion. The large-scale chart shows the metrics of in-degree, betweenness centrality and out-degree of approximately 50 network members, sorted on in-degree. We will discuss all three metrics on the following pages.
The inset chart is the compressed view of the entire set of members, sorted on in-degree. You can see that the distributions of in-degree and betweenness centrality are roughly in agreement, at least for the top 50 or so. Out-degree is a bit less closely related, although again, the top of the distribution is similar. We could assign a reasonable cut-off for people on whom we should concentrate where the red reference line lies.
In-degree: This image presents the network filtered to show only the nodes with a high in-degree (these were just chosen by eye, for illustration, but are probably close to the cut off that I suggested in the metrics chart above).Someone with a high in-degree, relative to others, receives more direct mentions and replies than others. This indicates status in network as other members want to communicate with them and reference them. Obviously, we would like to know more about these individuals as part of a communication strategy, as they can be expected to have influence on what is being said, and in particular, repeated.
Betweenness centrality formally measures the proportion of all of the shortest paths that exist between all of the nodes in the network on which a node sits.
A high betweenness centrality, then, means that that node is a “bridge” between or “connector” of many other nodes. If that node were removed, many other nodes would be either disconnected from each other or at least connected to others only by a greater number of links. Since diffusion of information is, other things being equal, easier the fewer links that need to be traversed, this property is very important to the structure of the network. High betweenness individuals can provide a short path to many players, and can prevent “structural holes” in the network, where groups of nodes are disconnected or only tenuously connected to the larger network.
Out-degree is simply how many tweets, mentions and replies you’ve sent. This is therefore clearly a measure of interest relative to diffusion, as we can expect high out-degree people to be active in spreading messages. The caveat here is that there is no filter for quality or the regard of the network involved, like there is, arguably, with in-degree. In the case of this network, it appears that there is at least a reasonable correspondence between out-degree and the other key metrics.
Two other metrics that we have not yet discussed involve the relationship between nodes.
Edge weight refers to the number of times two nodes link. In our case this would be repeated tweets, replies and mentions by one member to or involving another. This is usually taken as a proxy for the strength of the relationship on some dimension. The following view has been filtered to only include edges with weights of five or more. Note that these are relatively few and tend to involve our already-identified central players.
The colouring of the edges reflects the reciprocal or non-reciprocal nature of the relationship of connected nodes. A pair of reciprocally-related nodes is one where the parties are “following” each other (as you do on Twitter). These are represented in green. A reciprocal relationship is presumably stronger and is better for diffusion as a follower automatically receives the Twitter messages generated by those she follows.
Analysing tweet content: Because the software we used records the contents of each message, we can analyse the content qualitatively. This should be of considerable use in learning what caught participants’ interest, what issues were raised and so on at an event or in response to an issue or question.
We can understand even more about a network by looking at the temporal pattern of tweets. This chart shows a clear pattern of Twitter use in mid-afternoon and evening, probably corresponding to certain events. This indicates that there are times when the Twitter audience is more active and engaged and therefore open to messaging.
Twitter also records the geographic locations of some, but typically not all, members of a network. Time zones are more reliably recorded, which may be helpful in message planning.
This has been a demonstration of the potential for enhancing communications strategies, both during and after events, using mapping by social network analysis methods of Twitter traffic. Additional results can be obtained in mapping organizations’ Facebook pages. In the next post, we’ll look at how to evaluate the ongoing maintenance and growth of a social media network.