Measuring relevant networks

See our friends,
see the sights,
feel alright

Relevant net

I’ve already mentioned the paper by Bernardo A. Huberman, Daniel M. Romero and Fang Wu about Twitter’s social networks (pdf). The point that interested me most was the distinction between contacts and friends. Especially for people with a liberal following policy (i.e. that follow back once someone follows them on Twitter), their network of contacts is not very meaningful. But what is a meaningful network on Twitter? Huberman, Romero and Wu propose that there is a hidden network of friends that you just have to extract from your overall contact network. This hidden network is transaction-based: it’s the people you have replied to using the @syntax more than once. Because it is based on the choices you made, I’d prefer to call it the “relevant net” because it works similar to the “relevant set” known from marketing. This network consists of the few people from your total network that you find meaningful enough to actively address them.

This is my relevant net (calculated with “TwitterFriends”) consisting of all my Twitter contacts that I addressed with the @syntax more than once in my last 1,000 twitter messages (click to enlarge):


@furukama's relevant net on Twitter

@furukama's relevant net on Twitter

Inner circle

My relevant net consists of 92 nodes.  In my last 1,000 tweets, I addressed only 92 of my 1,400 contacts (6.6%) more than twice. But the replies are not equally distributed. There are many people I only addressed twice and there are some people I talked to very often. The above visualization shows the different layers of my relevant net. In the middle is the “inner circle” of my Twitter friends: a small number of people I addressed very often. The colors are important: blue are people I met in real life, orange are people I have not (yet) met and gray are bots or organizations. My inner circle is blue. This means that the people I talk to most often on Twitter are people I also know and talk to in real life.

Network analysis

The next step is taking a look at my Twitter friends and see how they are connected. Again, I did not use the formal notion of Twitter contacts, but the more specific notion of Twitter friends, i.e. people who talked to each other more than twice. To produce this network visualization, I removed myself, because all of the nodes in my relevant net are connected to me (that’s how they are defined) and so this does not add any information to the graph. Here are the conversations between members of my relevant net (click to enlarge):


Network of my Twitter friends

Network of @furukama's Twitter friends

Although this conversational network is not as dense as the network resulting from the following/followers network, it is not, as you would expect from what Huberman, Romero and Wu are writing, exactly a sparse network. Its density is 6,06%. This means, that of all 8,372 potential directed connections, 507 are realized. 8 nodes have received no replies at all. Without them, density would be 7,45%. Often, for directed valued graphs like this one (= graphs differentiating between incoming and outgoing ties) density is calculated as the average of the line weights (= number of replies between two persons). In this case, the density is 39. This means: the average number of replies from one of my Twitter friends to another is 39.

The distribution of the incoming @replies in my network looks like a power distribution, which means that this network should be a scale-free network. Few people in my network receive a large number of replies while a large number receives only few replies.

Distribution of incoming @replies

Distribution of incoming @replies

What does this all mean?

  • Social Media Measurement definitely should take a closer look at social network analysis. If you want to find out who are the influencers in a network like Twitter, looking at the numbers of followers could be misleading. @having has a large number of followers and is one of the nodes in my relevant net that gets the most replies. Is it an influencer? No, it isn’t. It doesn’t talk back.
  • Twitter should find a method allowing us to keep closer in touch with our relevant net. This small subset of all my contacts is often what’s interesting me the most. I’d love to have a tool that keeps me updated on them. But I don’t want to select my relevant net manually as it is possible with Twitbin-Groups. Above I’ve described a way, this can be done automatically.
  • This is a good example for the importance of relevance instead of total reach (e.g. number of followers, visitors, clicks) in social media. The actual exchance of messages, the dialogue that’s happening on Twitter is taking place in such relevant nets. Phenomena like these deserve much more attention by social media marketeers.

If you want to know your relevant net, you can calculate your Twitter friends with this application.

Share this post:
  • Digg
  • Facebook
  • Google
  • Furl
  • Ma.gnolia
  • MisterWong
  • NewsVine
  • TwitThis
  • Yigg
  • Pownce
  • Reddit
  • Slashdot
  • StumbleUpon
  • Wikio
  • LinkedIn
  • Spurl

Trackbacks/Pingbacks (2)

  1. TwitterFriends at viralmythen on Friday, December 12, 2008 at 8:05 pm

    […] ich tatsächlich im Gespräch bin. Man könnte auch “relevantes Netzwerk” (relevant net) dazu sagen. addthis_url = ‘’; […]

  2. Basic Thinking Blog | Wochenende on Saturday, December 13, 2008 at 2:35 pm

    […] – Benedikt Köhler hat die Applikation TwitterFriends veröffentlicht und erklärt seine Anwendung. Siehe auch measuring relevant networks […]