Twitfluence prototype calculation for measuring Twitter influence

icon

The prototype calculation of Twitfluence uses the data available form Twitter API to measure your Twitter influence and coolness. The basic technical specifications of the application is available, but I will also be supplying the basic information about how the algorithm works. The actual calculation is already online for beta users, and generally speaking, there are three major components that add up to the score: your followers, your mentions and retweets, and your lists, all accounted as ratios between you and others.

Followers

The strongest component of the calculation is the number of followers you have. In my opinion, your presence on Twitter and getting followers can be influenced by at least the following three major factors concerning you and your Twitter account:

  • Persona – how known you are. Measured by the number of followers you have, compared to your time on Twitter.
  • Engagement – how engaged you are. Measured by the number of followers you have, compared the number of people you follow; Measured by the number of followers you have, compared to the number of mentions and retweets you’ve made.
  • Wits – how smart and creative your tweets are. Measured by the number of followers you have compared to the total number of tweets you've made.

For this part, I gave the followers/following ratio the weight of 3, the followers/tweets a weight of 2 and the followers/time a weight of 1. The followers/(mentions + retweets) has a weight of 0.5 and works in the negative way, so people who bother other people get a bit of a minus to their followers result. Besides, those who are able to get the same number of followers without mentioning people, must have a small advantage.

(Needs to be upgraded with taking into account only your mentions and retweets of people who don’t follow you.)

Interaction (mentions, replies, retweets)

The second most important part of the calculation is the ratio between mentions and being mentioned, together with the number of retweets you get with the absolute "reach" of those retweets (measured in the number of people who follow people that retweeted you). A similar reach is also accounted in the mentions and replies. This component of the calculation uses only the data from the last month, also to make Twitfluence a bit dynamic for multiple calculations for a single user over time. To finalize this part, the total number of tweets in the last month also contributes a small score.

(Needs to be upgraded with unique reaches of your retweets and mentions. For now, it just adds them together.)

Lists

Twitter lists are getting used more and more, so they are also considered in the calculation. The number of lists you appear on, the number of people who follow those lists and the number of people, who follow lists you've created are the basic parameters for the calculation. This component adds only a small bit to the final score.

(Needs to be upgraded with unique reaches)

The basic ratio calculation

All ratios in the calculation are based on the same elementary formula, which looks like this:

Generic result = Sqrt(others / you) * Log10(modifier + 10)
Followers = Sqrt(followers / following) * Log10(followers + 10)
Mentioned = Sqrt(mentioned / mentions) * Log10(mentioned + 10)

I've decided to go for this architecture because of a number of reasons. F.i., the followers / following and other ratios are used to get an objective value for all Twitter users. This ratio gets square rooted so the differences between people are not so huge. The multiplication is there for adjustment, so people who have the same ratio and the absolute number are bigger, get more points. The logarithm is used to make this modifier of absolute number smaller, while + 10 is used so this number is always bigger than 1 (and the logarithmic function becomes more stable after the result 1: Log10(10) = 1). This means that the modifier for those who follow 10 people is around 1, 100 people around 2, 1000 around 3 etc.

Putting it together

The three major components currently have the following weight in the final score:

  • Followers: around 60%
  • Mentions and retweets: around 30%
  • Lists: around 10%

That's about it for now. I've tested the behavior with some real accounts (thanks for help @TejaSmeja and @jakasibicekaka), together with some projections, and it seems to be working quite OK. But the real test will happen after it analyzes results of actual people, which will allow real insight into the performance and objectivity. The Twitfluence will be online soon, and I will be asking you to help with testing the prototype. You also more than welcome to leave any kind of feedback about the calculation as I've described it.

Let's play.

Like what you read? Now tell the world!

A few more things you might find interesting:


Comment
written 1.8.2010 12:54 CET on chronolog
996 views   •   4 likes   •   Like   •   
date
date
date
date

Connect with :