About Tweetrank
How relevant are you?
We tried to calculate how relevant you are on Twitter. We took a look at all of Twitter, did some calculations and voila! Check out your score by entering your name in the box on the right or check out the complete ranking.
Filtered search
We used our ranking combined with Twitters search to make a cool filtered search. Using the slider you can filter out the tweets from users that are less relevant so you won't be bothered with all the chatter and spam.
Tweetrank is PageRank for Twitter
PageRank is the algorithm Google uses to rank their search result. We applied this algorithm to the entire Twitter social graph.
How about some math?
Twitter can be seen as one large directed graph where every user is a vertex and every follow-relationship is an edge. If user A follows user B, this means there is an edge from user A to user B. Everyone has incoming edges from all his followers and outgoing edges for everyone he follows.
To show you the formula we used, we need to make some definitions first.
- The rank of user
: 
- Followers of user :
Now, we define the rank of user as the sum of the rank of each one of
his followers, divided by the number of users they follow.
Because this would mean that every Twitter user is only interested in the updates
of the users they follow, we need a factor to compensate for that.
This can be done by introducing a randomization factor 
(which we chose to be equal to 0.9 for Tweetrank) and a variable equal to the
total number of Twitter users, 
. Combining these in the previous
formula leads to the following formula.
Recursion
Clearly, this formula is recursive. So how do we cope with that?
Initially we give every user the same rank of 
.
Then, the first computation of every is possible.
When applying the formula iteratively, the PageRank of every user converges to a stable value.
For Tweetrank, we stopped the algorithm after 50 iterations.
This means that if you have followers, that have followers, ..., that are followed by someone with a large 
, you also profit from that!
Wait, but this is quite static, right?
That's right. This doesn't take @mentions, RTs and other measures that make up Twitter-rep in account.
More so, the amount of every user sends out to the users he follows is equally weighed.
It is certainly possible to add weight for a relation if a user is retweeted, or @mentioned.
But, as the calculation of PageRank needs the entire Twitter social graph, this would require access to Twitter's firehose and more resources than we currently have access to.
The Numbers
The Twitter social graph is large. We have crawled 50 million users with 1.8 billion relations. This resulted in a dataset of 8GB in size. Obtaining this dataset took 3 weeks at a rate of 100k API calls per hour.
Not everyone is in the list
Because Twitter doesn't let us look at users who have protected their tweets, not everyone is in our ranking. Also, as crawling the entire Twitter user base takes a couple of weeks, data will always be old. The data that is used on this website was crawled a couple of weeks ago so new users might not be in there. Also, users who have gained a lot of rep in the meantime may still be ranked low.
Any questions?
Just mention @tweetrankme on Twitter with your question and we'll try to get back to you!
About us
Who are we?
We are @pnoordhuis and @michielheijkoop and we decided to do this project for a course on web technology at the University of Groningen.
Thanks!
To our professor, Alexander Lazovik, for helping us with this project, to Amazon for giving us an educational grant to use their cloud computing service and to Twitter for allowing us to crawl their data. And of course to you for checking out our work!
