Data Labs: Mapping out the UK’s Political Network

With the election less than two months away, Twitter is full of election coverage and political debate.

Mentions of the major parties have increased to over 50,000 Tweets per day, as referenced in Matt’s earlier blog post.

But who is weighing in on the debate, what issues do they care about, and is it a unified conversation or a battle on several fronts?

If we take a deeper look at the interactions among commentators, we can find the distinct communities that make up the political Twittersphere.

---

The data

I used Brandwatch Analytics to listen out for people mentioning UK political parties on Twitter in the run-up to the election.

After filtering out organizational accounts, I ended up with 29,138 Twitter users in the UK who’ve mentioned politics in the past three months.

How can we find social connections among the accounts? One approach would be to use the follower network. However, following is cheap. It does not necessarily mean that people know each other, interact, or even read each other’s Tweets.

Instead, I used retweets and @mentions, collected over the past year as part of PeerIndex’s influencer analytics pipeline. These are the strong links in the social graph. Compared to following, they’re better indicators of whose content gets noticed by whom. After all, we can only hope that people read Tweets before they retweet!

---

Detecting communities

I divided the graph into communities using the Louvain method, which finds clusters of nodes that interact more strongly with each other than with neighbouring clusters. There are seven major communities with more than 500 Twitter users apiece.

What do the communities mean? We can get a sense of what sets each community apart by looking at how people describe themselves in their Twitter bios.

Not everybody has a Twitter bio that’s informative about their occupation, interests, or political affiliation, but relative frequencies of words in the different communities give us some idea of what sets them apart.

For each community, I found the words from Twitter bios that were best predicted by community membership.

For example, Community A most closely matches the distribution of the word writer.

A tag-cloud is a good way of representing these labels. This method automatically leaves out words that are common across all communities, and instead highlights what sets the communities apart.

 

The largest community contains more than a third of the network.

Its members tend to describe themselves with words like media, writer, editor, digital, and geek.

The names of the main political parties in England fall into three communities.

One includes the current coalition parties and UKIP, whereas in the other two the word labour is more common.

Of the two left-leaning communities, one has more of a focus on health, with a stronger association to socialism and the Green Party. The second Labour community contains more people in elected positions and fewer who describe themselves as activists.

Scotland and Northern Ireland both have their own distinctive clusters.

Sinn Fein and Ulster Unionists are both present in the Northern Ireland community, whereas people in the Scotland community frequently identify themselves with the independence debate.

The “Yes” campaign dominated on Twitter at the time of the referendum, and vote yes is still prominent in the Twitter bios of the Scotland community.

Finally, there is a community interested in animals and the environment.

The distribution of green dots in the force-directed network diagram above shows that there are several prominent sub-communities, which could be related to the variety of issues that this community identifies with, including animal cruelty, energy, and conservation.

 

---

Flow of political messages

The automated community detection above is based on any retweets or mentions – not just Tweets about politics.

How does information about politics flow between communities?

To answer that question, I analyzed retweets about politics and the election from the past 3 months. The sizes of the circles are proportional to the numbers of people in the communities.

The arrows represent volume of retweets, going from the community where the message originated to the community that retweeted it.

As we might expect from the clustering algorithm, each community retweets more political content from itself than from any other community.

The proportion of retweets coming from within the community rather than outside ranges from 75% for the Scotland community (most “independent”) down to 25% for the environment community (listens most to the outside when it comes to politics).

The most frequent retweets across community boundaries are among the main Westminster parties, with especially strong links between the two left-wing communities.

The socialist, conservative, and Scottish clusters get the most retweets per capita. The media and environment communities are more likely to retweet messages from the party-political communities than vice versa.

---

Concluding points

• Political commentators in the UK fall into seven major communities, based on how people interact on Twitter.
• Communities are defined by regional conversations (e.g. Scotland, Northern Ireland), political allegiances, and topics of interest (e.g. digital media, wildlife).
• Messages about the election are mostly retweeted inside the community where they originated, although the party-political communities get more outside engagement.

---