In this blog post we provide some insight into the the Twitter account of @PeterJukes, a journalist, English author, screenwriter, playwright, literary critic and blogger. We follow a similar methodology to that of previous work where analysed the Twitter account of Paul Mason.
Peter, is a prolific Twitter user with 199 thousand tweets, and 23 thousand followers. So we thought that his account would make for some good analysis. We used the Fruchterman Reingold graph layout algorithm produced using NodeXL which displays the Twitter network of Peter Jukes. The tiny boxes represent nodes and this case these are Twitter users, and the lines represent ‘edges’ which are the lines connecting Twitter users.
The network graph below displays all the Twitter accounts that have engaged with the Twitter handle ‘@PeterJukes‘ over the previous seven or so days.
Figure 1 – Network graph 1 using the Fruchterman Reingold graph layout algorithm
The image above displays the Twitter network of the ‘@PeterJukes’ Twitter handle. See the NodeXL graph gallery for the full image, alongside a detailed report, and an interactive explorer.
We can use the interactive explorer functionality to zoom into the network:
Figure 2 – Using the interactive explorer to zoom into network of ‘@PeterJukes’
In the figure above we have hovered over the ‘@PeterJukes‘ Twitter handle which is towards the centre of the network, and the accounts closest to that of Peter’s account indicate strong connections i.e., they may be accounts that are frequently mentioned or those which engage with Peter’st tweets the most. The connections between Twitter users are directed, for instance, Donald Trump appears in the graph as he has been tagged in tweets, similarly to that of CNN.
This is one way of laying out Twitter networks, but one of our favorite methods is by using NodeXL to select the Harel-Koren Fast Multiscale algorithm to layout the graphs. This finds groups and divisions within a Twitter network, as it clusters Twitter topics based on the same network into different groups. Discussions on Twitter can form into a number of different groups, see figure 2 below for 6 types of networks that can form.
Figure 3 – Six Types of Twitter Network (see Smith, Rainie, Shneiderman, & Himelboim, 2014)
As show in the figure above different topics on social media can have contrasting network patterns. For instance in the polarized crowd discussion one set of users may talk about Donald Trump and others about Hilary Clinton, in the unified crowd users may talk about different aspects of the election, and in brand clusters people may offer an opinion related to the election without being connected to one another and without mentioning each other.
In a community cluster a group of users may talk about the different news articles surrounding Hilary Clinton. Broadcast networks are typically found when analysing news accounts as these disseminate news which is retweeted by a large amount of users. We can think of support networks as those accounts which reply to a large number of accounts, we can think of the customer support of a bank which may reply to a large amount of Twitter users.
Figure 4 – Network graph 2 using the Harel-Koren Fast Multiscale algorithm
The network graph is made up of several groups of Twitter users. Group 1 displays the Twitter audience of @PeterJukes, which is has elements of Broadcast Network which indicates users are retweeting Peter, but also that of a Support Network, as Peter is replying to and engaging with other Twitter users.
The top left hand side of each group is labeled e.g. G1 refers to ‘Group 1’ and G2 refers to ‘Group 2’ and so forth, and the keywords relate to the most frequently occurring per each group. NodeXL network graph reports also produce a number of other metrics such as the most frequently shared URLs, Domains, Hashtags, Words, Word Pairs, Replied-To, Mentioned Users, and most frequent tweeters These metrics are produced overall and also by group of Twitter users. By looking at different metrics associated with different groups (G1, G2, G3 etc) you can see the different topics that users may be talking about.
The full report can be accessed here, and there are enough analytics in the report to write an entire PhD thesis, but for the purposes of brevity, in this post we can examine some of the most frequently occurring co-words (words that appear together), keywords, and hashtags, and articles to get a sense of what types of topics have dominated Peter’s network over the last week.
Top 4 URLs
1. A link to Peter’s tweet which links to a New York Times article related to Obama and Trump
2. A link to a Guardian article titled ‘All Out War; The Brexit Club; The Bad Boys of Brexit review – rollicking referendum recollections’
3. A link to a Who What Why article titled ‘ A Dark View From Flyover Country ‘
4. A link to a Politics.co.uk article titled ‘The British press is an enemy of free speech’
Examining the network graph and the most frequently occurring hashtags, URLs, and words, we can conclude from the data that a popular topic among the network over the past week was related to that of article related to Obamas view’s towards Donald Trump, and specifically this tweet which was sent by Peter promoting the article. It is important to note that these graphs are topical and change on a week by week basis, we would recommended browsing the full report on the NodeXL graph gallery.
Just for a bit of fun, and because we are interested to examine the days of the week and times journalists tend to tweet. We used Twitonomy, to examine the average time and day Peter tweets, as shown in figure 5 below.
Figure 5 – Days of the week and time tweeted
It looks as though Saturday is the most active day, with 9pm as the time of choice for sending tweets.
Here at Sonic Social Media we like journalists, and so we offer completely free of charge analysis of accounts which belong to journalists. You can complete this form and we can have something back to you within 24 hours.
Want to learn more? Check out this NodeXL and Social Network Analysis workshop at Leeds Becket University which will cover all of the methods used in this post in much depth.