Ad verba per numeros

On user influence and algorithmic ways to compute it

Sunday, May 8, 2011, 04:10 PM
It seems I just find the time to blog once in a blue moon; anyway, I had some interesting discussion with master's students last week and I think they deserve a post (albeit a short one).

One of the topics raised during those discussions was the "influence" or "impact" of users in online social networks. In fact, one of the proposed exercises consisted of studying, implementing, and analysing the so-called velocity-acceleration method by Organic, Inc. ([1] and [2]).

Thus, the discussion revolved around what "influence" really is, measures for the different interpretations of influence (if any), and algorithms to compute them.

Needless to say, different perceptions for influence arose:

  • get other people to accept your ideas and spread them (e.g. getting retweets in Twitter or Likes in Facebook);
  • get people to consume your contents or the contents you promote (e.g. getting people to click in the URLs you publish);
  • or get people to behave in a certain way in real world (e.g. buying a product, attending a concert or voting a given candidate).
The further we move from the online to the offline world the more difficult it is to measure the impact of online actions and, thus, we decided to limit ourselves to online behaviors (which, by the way, are still far from being solved). Thus, retweets and URL clicks were considered the most interesting metrics in order to measure the "influence" of users.

With regards to algorithms to compute one or another measure of "influence" or "impact" there are a number of them for Twitter and other online social networks, such as:

In addition to them there are a number of previous works that deserve a reading and, moreover, could be applied to online social networks:
So, the morale of the story is two-fold:

First, influence is an elusive concept so you'll need to determine the kind of impact you're interested in and the way to measure it.

Second, there can exist an algorithmic way to compute a proxy measure for your "flavor" of influence so give a try to previous methods before going after a new silver bullet.

As usual, if you want to add something to the discussion just e-mail me or contat me at Twitter (@PFCdgayo).


Back Next