Twitter is a great place to share ideas, make connections, let off some steam or get some support. It's also full of bitter arguments, bots and mobs, hashtags that are deliberately gamed for attention and misinformation that's deliberately spread. Looking at the network of a Twitter conversation is one way to tell what kind of conversation it is.
Working with the Social Media Research Foundation to build visualisations with NodeXL (a free Excel add-on), the Pew Research Center found six main types of conversation.
Political conversations are often divided between two polarised groups who talk among themselves, but not to the other group, linking to different sites, using different hashtags and clustering around different discussion leaders: that's the polarised crowd.
Tight crowds are a very different pattern: these are groups where almost everyone talks to almost everyone else about a topic, tossing around ideas, suggesting solutions and sharing information.
Lots of different people tweeting about the same thing but not talking to each other - discussing a celebrity or a brand or a news story - looks different again, with lots of disconnected participants and hardly any actual networks forming.
Community clusters are somewhere between the two: instead of one big tight crowd there are multiple smaller crowds of people talking to each other but not talking to those other crowds. Those clusters might be existing networks of people (think six degrees of separation) or they might crystallise around different takes on a topic from different news sites or commentators.
For a news story breaking from one big media site, the broadcast network of people repeating and retweeting without interacting with each other - with maybe some discussion groups of people springing up to have a conversation with each other, but not getting responses from the many people spreading the news.
That hub and spoke pattern is different to a company twitter account trying to handle complaints and questions where the outward spokes tend not to spark discussion groups (unless a particularly good or bad response goes viral).
I asked Marc Smith from the Social Media Foundation what kind of patterns my Twitter discussions formed and how that compared to some more notable tweeters, particularly those covering political issues in the UK. He used NodeXL with the free public Twitter API that grabs up to 18,00 tweets for a single query. The Twitter data recipe for doing this is on the SMF site, and there's a video guide to analysing tweets yourself here.
The hashtag didn't look like a two-group polarised crowd; instead there were multiple clusters, broadcast networks and discussion groups. I was glad to see that the same was true of my own Twitter account (snapshotted on a day last October when I was discussing politics, Microsoft Ignite, Azure, security, the lack of capex investment in IBM's cloud infrastructure, the importance of deleting PII that your data doesn't need, the NodeJS community, and various other topics).
The other tweeters had patterns that all looked very similar to each other, and a little different from mine because their larger networks mean the patterns are on a much larger scale - but it was the same underlying pattern.
"The main network patterns I see are hub and spoke "broadcast" patterns - lots of rings and rings around rings created when many accounts retweet a central account (and sometimes get lots of retweets themselves, creating the outer rings)," Smith told me.
"At the centre of each of these clusters is the hub account, at the top of each cluster is the list of hashtags that define that cluster. Often clusters do not have alignment with other clusters."
Using NodeXL to analyse tweets isn't something you can do in real time on your own PC; it can take an hour or more for the data to get imported and processed. In future, NodeXL will be available for Power BI, and that should process data rather faster in the cloud.
Either way, if you want to understand what's going on with a particular Twitter hashtag that's becoming influential, taking a look at the shape of the Twitter network behind it can reveal a lot.