Nothing wrong with recursion. You can keep re-applying that and normalize/rebalance the 'scores'. This is what one Mr. Brin did with one Mr. Page for web search. It really worked out well for them.
I created an influencers analytics solution some years ago (tribalytics.com), and Social Network Analysis (SNA) applied to people's followers got us much better results: Instead of analyzing a twitter user's bio, we would also cluster their followers into communities (according to how they follow each other), then summarize each community's characteristics. This gave us multiple accurate segments allowing to qualify a person
e.g.:
* 35% marketing and social media professional
* 10% foodie
* 15% kayak fan
* etc...
I'm not active anymore on this solution though, as that analysis was quite expensive and too complex for what our target customers marketers were looking for...
Also who are the big leaders in the influencer search tool space these days?
Indeed, she participated in tribalytics for some time.
> Also who are the big leaders in the influencer search tool space these days?
I'm a bit rusty as the landscape evolves quickly, and I'm not much active anymore on the influence marketing sphere. Two actors which look relevant to me:
* getlittlebird.com
* traackr.com
There are both tweet-level and account-level signals. But some of those signals can be artificially inflated -- follower counts, favorites, etc. One of the highest signal features I focused on was whether or not you were RT'd or @'d by another influential account. Of course, this is a somewhat-recursive problem which made it difficult to wrangle with.
I eventually got something to work, but it was only possible because I had access to the Twitter firehouse and through the manual weighting of features.