A few weeks ago, I spent a long day throwing together a new Twitter application, “twinfluence.com“.  In about 8-10 hours coding PHP and MySQL with the Twitter API, I launched the site in time to present it at a BarCamp.  The primary idea behind it was that simply counting followers was an insufficient way to measure influence via Twitter.

Counting followers treats all followers as equal, and assumes that any tweet message ends with one’s followers - which we all intuit as wrong.  Ideally, we would methodically map out the entire network of Twitter and calculate some rigorous social network analysis metrics like clustering, betweenness, equivalence, etc. to really understand the structure of the Twitterverse and the role any particular Twitterer plays in it.  Sadly, the limitations of internet communication and the Twitter API make that completely impossible for a “real time” twitter analyzer like twinfluence.  I was forced to limit twinfluence’s real-time “crawling” and indexing to the second-order network - followers of followers (and still have to cache those with 30,000+ followers for offline crawling and analyzing).  Again, it’s an arbitrary cutoff (Josh Carrico’s blog discusses this nicely, and of course retweeted messages can travel further than that), but extending the horizon to the second order network has helped change the discussion of influence to include whom one’s followers are, and how much audience each brings to the party.

To quote the intro on twinfluence.com’s “about” page:

Imagine Twitterer1, who has 10,000 followers - most of which are bots and inactives with no followers of their own. Now imagine Twitterer2, who only has 10 followers - but each of them has 5,000 followers. Who has the most real “influence?” Twitterer2, of course.

I also wanted to make certain that the metrics were all transparent and well-defined, and some global trend statistics were collected so everyone could get a better idea of what should be the “baseline” expectation as one’s twitter network grows.  Twitter Grader (probably the most comparable Twitter tool out there) is an impressive app, but it doesn’t live up to my expectations on those counts.

Twinfluence ranks on Reach (second-order followers), Velocity (how quickly reach is built), Social Capital (average followers’ network size), and Centralization (how much the network relies on a few high-profile followers).  Detailed definitions are available on the twinfluence.com “about” page.  In two weeks, over 7,000 twitterers have been analyzed, and we’ve learned that as a twitterer’s network grows - (1) they tend to pick up a few high-profile followers, but over time new followers tend to have smaller and smaller networks; (2), their networks decentralize over time; and (3) velocity picks up (snowballs) as network size grows.

Twinfluence has gotten a lot of attention, discussion, and worthy critique.  Some (of many) of the blogs that have picked up the discussion:

• http://mashable.com/2008/10/23/how-to-improve-twinfluence-and-twitter-grade/
• http://www.inquisitr.com/5393/twinfluence/
• http://theseldomseenkid.blogspot.com/2008/10/twinfluence.html
• http://www.killerstartups.com/Web-App-Tools/twinfluence-com-twitter-influence-analyzer
• http://chocolate-fix.blogspot.com/2008/10/sphere-of-twinfulence.html
• http://www.thisisherd.com/2008/10/twitter-why-you-ignore-it-at-your-peril.html

The latest conversations I’ve had on twitter is that the social capital metric is skewed toward twitterers with small networks.  We can probably improve the social capital metric, make it more realistic and useful by devaluing the contribution of followers who themselves have thousands of followers (those probably don’t contribute real “capital” anyway, since they probably aren’t listening).  When I hash out a new, workable algorithm we’ll release a new version of Twinfluence.  My thanks for suggestions/feedback by ScottAllen, Macartisan, Sigepjedi,  CCSeed, Yannleroux, Mlaine, KevinLyons, PatchWalker, RepeatNone and others.

| Permalink | Trackback