The 70+ Parts of the Twitter Grader Algorithm

Social Media

by Kenny Hyder

By now, most of the hyper-active twit­ter users are aware of the twit­ter pro­file ran­king tool twitter.grader.com. Some of us even check it daily. (Just mes­sin with you)

It “gra­des” your twit­ter pro­file on a scale of 100 based on:

  • The num­ber of follo­wers you have
  • The power of this net­work of followers
  • The pace of your updates
  • The com­ple­te­ness of your profile
  • …a few others
  • At least that’s what they tell you ;) Of course, I wan­ted to know how it really wor­ked. Here’s what I’ve found so far.

    Here is a graph of me and some of my reci­pro­cal follo­wers sor­ted by num­ber of upda­tes:
    Reciprocal Followers

    These num­bers are as of this mor­ning, 10–20 bet­ween 9:00 — 9:15 AM PST. As you can see, the first 7 on the list are all extre­mely active users, with sig­ni­fi­cant follo­wings. Their sco­res vary from 94 — 98.9. I deci­ded not to inc­lude anyone with a 100 score in my study, because those users tend to have such large follo­wer num­bers, that the results I feel, would be ske­wed. The last 3 on the list are users, inc­lu­ding myself, with sco­res in the middle range, var­ying from 57 — 63. Follo­wer counts for these users are all under 100.

    Now, there is obviously a corre­la­tion bet­ween num­ber of follo­wers and twit­ter score. I would liken this to num­ber of qua­lity inbound links to a web­site, and page rank. But there are some things to take note of.

    I first thought that it would corre­late that the ratio of follo­wers to peo­ple you follow should sig­ni­fi­cantly affect your score. But, as you can see with NeO­Blog, this is not the case. He is follo­wing more than 2 times the amount of peo­ple that are follo­wing him, yet his score remains at the top with a 97.5, and con­ti­nues to climb. Whe­reas chris­win­field, on the other hand, has over 6 times as many follo­wers than he is follo­wing, and only main­tains a mere 1.4 score lead. We see a simi­lar case with the epic duel of mar­tin­bow­ling vs. oil­man. Mar­tin­bow­ling only main­tains a 1.6÷1 ratio of followers/following whilst oil­man has a stun­ning 4/1 ratio yet only leads by a .6 points on twit­ter gra­der. So I had to aban­don this logic.

    I next deci­ded to look into qua­lity of the net­work of follo­wers. For this I more clo­sely exa­mi­ned the last 3 on the list, because the first 7 have many over­laps in net­works. Robert­pal­mer, kennyhy­der, and aus­tin­cur­tis all have simi­lar follo­wer num­bers with few over­laps. This was an inte­res­ting study, and one that is hard to not be bia­sed on! ;) I’m an SEO and follo­wed by mainly the SEO and inter­net mar­ke­ting com­mu­nity. Aus­tin­cur­tis is a pro­fes­sio­nal pho­to­grapher and desig­ner, and mainly follo­wed by other pro­fes­sio­nal pho­to­graphers. And robert­pal­mer is an author at www.tuaw.com and a graphic desig­ner, and mostly follo­wed by other blog­gers and desig­ners. Because both Aus­tin and Robert were per­so­nal friends of mine before I knew them pro­fes­sio­nally, and I can’t just say “my follo­wing is bet­ter” I deci­ded to look at the num­bers. Without going too deep, I deci­ded to look at the num­ber of follo­wers for each user, who have a follo­wer count above 1000. The score? robert­pal­mer: 14, aus­tin­cur­tis: 5, kennyhy­der: 14. With the robert­pal­mer and my count tied, and aus­tin­cur­tis at a sig­ni­fi­cantly lower count, yet all 3 users still so close, I deci­ded to play around a bit. So I went and bloc­ked all of the spammy users who were follo­wing me at the moment. (no pic­ture, follo­wing 1000 but only 5 follo­wers, etc..) The result? MY SCORE DROPPED! I laughed, obviously twit­ter gra­der doesn’t con­si­der the qua­lity of the accounts follo­wing you.

    After this, I deci­ded to follow @grader, based on the “…a few other things” line, and the fact that every time I chec­ked my score, at the bot­tom it said “kennyhy­der is not follo­wing @grader yet”. I thought this may do 2 things. 1. Pos­sibly boost my score a point & 2. I thought it would help with the fre­quency of crawl rate on my pro­file.. It did neither.

    I was star­ting to think that the algo­rithm was simply nothing except how many follo­wers you have. The last thing I had to try was the “pace of your upda­tes”. So I star­ted twee­ting like mad. I star­ted twee­ting @ peo­ple, twee­ting when I pos­ted on my blog, twee­ting when I was pla­ying poker, twee­ting everything.. And soon, the big lead that I once had on aus­tin­cur­tis tur­ned into a small lead. And soon, into his lead. I domi­na­ted him on tweets, and even lead on follo­wers, and was follo­wing less peo­ple, but his score pop­ped up. And then I saw something new, “What about your follo­wer to update ratio?”

    So to test this, (and get back my lead on aus­tin­cur­tis) I star­ted follo­wing new friends. My follo­wing count sur­pas­sed my follo­wer count, yes, but we already deter­mi­ned that this doesn’t mat­ter. I follo­wed more and more tweeps, and pic­ked up follo­wers along the way. I did this on fri­days because I don’t tend to spend much time on twit­ter over the wee­kends, so my follo­wer count would grow, and my update count would stay the same. This wor­ked. In further tes­ting, I would tweet a lot without adding friends, and my score would dip. Aha! This is the sup­ple­ment to main piece of the algo­rithm. This is the “con­tent is king” piece!

    So obviously, the more follo­wers you have, the bet­ter your score will be, but it is nice to see (at least for an seo) that this isn’t the only thing taken into consideration!

    Peo­ple men­tio­ned in this post:

  • Mar­tin Bow­ling aka @martinbowling
  • Todd Frie­sen aka @oilman
  • Dave Sny­der aka @davesnyder
  • Kate Morris aka @katemorris
  • David Brown aka @NeOBlog
  • Chris Win­field aka @chriswinfield
  • Frank Watson aka @AussieWebmaster
  • Robert Pal­mer aka @robertpalmer
  • Kenny Hyder aka @kennyhyder
  • Aus­tin Cur­tis aka @austincurtis
  • PS: If you liked this post, FOLLOW ME :) , and then check out this one by David Brown.

    { 9 comments }

    Kate Morris October 20, 2008 at 12:36 pm

    Thanks for the mention!! Very interesting piece!

    Austin Curtis October 20, 2008 at 12:42 pm

    ahhhh! Nicely done my friend. Nicely done. We’ll have to see if Twitter rank actually ever amounts to anything. Either way, it proves that we’ll go to extensive lengths even for imaginary points – hehe… Thanks for the post man!

    ps: could you change my anchor text to “Santa Barbara Wedding Photographer”? Thanks. HAHA!

    Andy Beard October 20, 2008 at 12:47 pm

    Test to see what happens when you follow a few political tweeters who autofollow back ;)

    admin October 20, 2008 at 12:58 pm

    Heh, yeah, my guess is it won’t do much, but it will at least be a follow right? :)

    Dharmesh Shah October 20, 2008 at 4:05 pm

    Thanks for the thorough analysis of the Twitter Grader algorithm.

    I learned a few things myself (I’m the developer). :)

    The application is still in alpha so it’ll be interesting to reconduct the analysis once we go into beta.

    We’re still tweaking the algorithm (and clearly, still have some work to do).

    -Dharmesh (@onstartups)

    admin October 20, 2008 at 7:22 pm

    Heh, thanks Dharmesh!

    There’s always tweaking when an algorithm is involved!

    I look forward to updates and seeing how things evolve!

    Brian October 27, 2008 at 3:15 pm

    OCD is awesome ;) Thanks for the analysis, now back to Twitter to work on my imaginary points :)

    Austin November 2, 2008 at 10:12 pm

    Kenny – check out this interesting tweep: http://twitter.com/uncle_bob

    He’s following 353, w/ 190 followers, and only 63 updates. Grade = 84.

    I think you’re right about the updates-to-followers ratio.

    PEACE

    Arthur Charles Van Wyk June 21, 2009 at 1:25 pm

    This was great education. Thank you very much. You took the time to test and the time to share. NOt all of us have the time to do what you did, but I know many that will appreciate the findings of your succinct study.

    Thanx again

    Previous post:

    Next post: