Twitter Grader Follows @‘s

Social Media

by Kenny Hyder

Just 2 days ago, I pos­ted about the update to the Twit­ter Gra­der algorithm.

In the post I cove­red how the algo update now counts follo­wer to follo­wing ratios in their sco­ring and some other cool stuff. I also made a recom­men­da­tion for the next update: “My recom­men­da­tion for the next algo update? Figure out a way to incor­po­rate user inte­rac­tion (@’s, ret­weets, replys) After all, it’s all about the conversation ;)”

Later that night, I recei­ved yet another com­ment from Dhar­mesh Shah, the deve­lo­per behind the Twit­ter Gra­der algo. He said:

Thanks for the second round of thought­ful analysis.

The algo­rithm was indeed upda­ted this wee­kend (and I think it’s “bet­ter”).
Quick note: We have 200,000+ uni­que pro­fi­les that have been gra­ded. In a way, this is good, but the flip side is that it takes some time to “nor­ma­lize” the data­base of gra­des. We’ve got sig­ni­fi­cant ser­ver resour­ces powe­ring the soft­ware, but I tend to “trickle-in” the recal­cu­la­tions. That’s a long-winded way of saying: It’s going to take some time for the actual gra­des to com­ple­tely reflect the upda­ted algorithm.

Making pro­gress (I think). The point about trac­king ret­weets and the “qua­lity” of con­ver­sa­tions for a given user pro­file is a really good one. I’ve been thin­king about that one and will try to incor­po­rate something like this in a future update.


Well, it looks like the future update came soo­ner than expec­ted! 2 days is a pretty good turn-around time if you ask me! Today, lots of users are noti­cing a boost in their Twit­ter grade because of what I believe to be an incor­po­ra­tion into the algo of user-interaction.

Here are a few things to look at:

Just for good mea­sure, here are the charts, even though they are of less sig­ni­fi­cance this round..

Upda­ted Chart:

Recip Grader Followers








And the Pre­vious Chart:








Now, take a look at mar­tin­bow­ling, kate­mo­rris, and neo­blog. All of these users saw somewhat sig­ni­fi­cant jumps in their gra­des, without sig­ni­fi­cant inc­rea­ses in follo­wers or other acti­vity in upda­tes or follo­wing num­bers over the past 2 days.

BUT, take a look at these screenshots: (click for popup) mar­tin­bow­ling kate­mo­rris neo­blog

All of these users have a high volume of inco­ming @‘s, part of the user-interaction recom­men­da­tion that Dhar­mesh said would be incorporated!

But there’s more.. I also chec­ked @gary­vee’s score, who I have been moni­to­ring over the past few months. He has 18,352 follo­wers, follows 2,134, and 3,306 upda­tes. The past cou­ple months his score has varied from 99.9 to 100.1. But is noto­rious for not repl­ying to inco­ming @‘s that he recei­ves. Today, after the update, his score is down to a 99.7. Con­sis­tent with this trend is @chris­win­field, who also saw a slight drop in score today. Chris and I actually know each other apart from twit­ter, and he’s a great guy.. But, he gets a TON of @‘s because of all of the twit­ter polls and ques­tions that he does, but doesn’t @ back at nearly the volume that he gets them in. (sorry chris ;))

This leads me to believe that Twit­ter Gra­der has incor­po­ra­ted in an update to their algo­rithm that accounts for inco­ming and out­going @‘s on your twit­ter account, like I had men­tio­ned in my last post.

If you find this to be true, leave a com­ment so I can check out your pro­file! And remem­ber tweeps, its all about the conversation!


Austin November 5, 2008 at 1:39 pm

sweet! All my @ing has finally paid off! hahaha

David Leonhardt November 5, 2008 at 5:05 pm

Well, just because I like being a rebel, I’ll take an opposing view. Although some @ comments are worthwhile, one really has to be careful to provide context…otherwise you end up filling up all your followers’ home pages with unintelligible bits of conversations. After a couple weeks on Twitter, I have already deleted some people because all they do is add noise with no meaning to my page.

In 140 words, tweets really should be able to stand on their own. If I was in charge of grading, I would give more weight to a person who is able to create tweets that do not result in a lot of @ noise.

admin November 5, 2008 at 8:47 pm

@david – that’s an interesting view, and I definitely see your point, the problem is, that it is very easy to say things that stand alone, and “spam” a twitter profile. Creating conversation that other people want to get involved in identifies you as a human, and an interesting one at that. User interaction and collaboration is the heartbeat of social media on the web.

Dharmesh Shah November 6, 2008 at 9:45 pm

Kenny: Thanks again for all of your time and effort. I think we’re at a point now where the grading algorithm is relatively decent. Always more work that can be done.

@david: I’m with Kenny on this one. I just don’t think that having stand-alone tweets is really a signal of quality that should be weighted highly. These kinds of tweets are nearly indistinguishable from spam.

LPetrides November 7, 2008 at 1:59 am

Just curious: Earlier on, it also seemed that part of algo included relative strength of network (how many followers the people you followed or followed you have). As someone who has single-handedly introduced twitter to dozens of people who have subsequently created accounts (but are still newbies themselves), is my score lower than if I just followed or am followed by the “biggies” scoble etc.?

Dave L November 7, 2008 at 9:24 am

So this would seem to mean that moving DMs over to @’s would increase your grade.

Michael Russell November 8, 2008 at 3:18 pm

Really appreciate your insights into the mysterious sausage grinder that is the Twitter Grader algorithm.

Lord knows I’m not complainin’…I’m just thrilled to find myself up in the 90-somethings with @laelaporte, @garyvee, @chrisbrogan and other notables.

Chris Blair June 28, 2009 at 9:47 pm

I know there were alot of issues with Twitter changing the way @replies worked with how they were showing in the public timelines, and it could be an interesting investigation to look at what “client” the people who go up/down from the new algo are/were using (eg are some clients recording your @’s with the twitter grader system better than others?) /// Another thing to monitor could be is a one word @reply saying “thanks” as valuable and worth as much as a full 140 character response. Is a valuable response one that gets an @response back to your @response?

Previous post:

Next post: