notes for interpreting Twitter network maps

After looking at a couple of hundred maps made from data-sets of tweets, intuitively see certain arrangements – courtesy of a highly tweakable algorithm – as indicating communities of belief, arguments between communities, divides, areas of accord and areas of dispute. 

The placement of prominent accounts in a map indicates their role in a dispute – antagonist,or target.  The overall shape of the discourse is  telltale too, because across  many maps (and many subjects) the same  factions, communities, clusters, tend to manifest, their presence inflated or diminished, depending on their scale, influence, and interest.

As a starting point, a map in Gephi of a data-set constituting a set of tweets from accounts, all containing a particular term or set of terms, looks like this:

The map has nodes but no connections.

But if one of the tweets in the data-set is a reply to another tweet, the connection’s mapped too.

The algorithm that arranges the nodes, introduces a force, a simulated gravity, that pulls nodes towards one another if they’re connected.

A map with more connections (a connection may be a retweet, a reply, a quote-tweet, or a mention), ‘communities’ become apparent.

The attractive force supplied by the algorithm between connected nodes enhances the clustering effect, making communities more visible.

Nodes can be resized based on how many connections they have, or their ‘degree’.

For a straightforward network like this a statistical analysis can easily pick out the three communities and colour the nodes, because nodes are only connected with other nodes in the same community.

It’s more difficult to identify communities when the map has thousands of nodes, and tens or hundreds of thousands of connections: that is, thousands of accounts are tweeting using the search terms used to collect the data-set, and many are replying to one another, or retweeting, quote-tweeting or mentioning one another. Also, communities don’t only communicate amongst themselves: there’s plenty of cross-fire between communities in virulent disaccord.

For the map above, a different method’s used to show distinct political communities, and intercourse between them. The method is based on colouring the lines that represent relations between tweets according to the type of connection. . The difference between these types of response can be used to find ‘communities of belief’, sets of accounts with content that tends to agree, in terms of world-view and ideology.