Learning about Twitter Trends!

I got into a pretty... annoying discussion (I'll admit) about Twitter Trends. Seems some folks are upset at Twitter, accusing them of censoring out some Paris tweets in favor of more "progressive" conversations such as #MuslimsAreNotTerrorists. I joined the conversation with the intent on teaching people something new. For example, they kept talking about the "algorithm" like they didn't fully understand how it worked. As predicted, they just argued rather than listened.

How the Trending Topics algorithm works:

In general, (because not everything is known about the algorithm, and it's constantly changing), the "Trending Topics" on Twitter work on a foundation of Acceleration, not Velocity. If you're in a car traveling at 100mph for 24 hours straight, what was the change in acceleration over the last 24 hour period? 0mph right? Now, if you were in a car traveling at 5mph, and accelerated to 50mph - your change delta is +45mph right?

In that example, even though you're going much slower than the person traveling at 100mph, you *accelerated* more than that person did over the same period of time. So in that analogy, you would Trend higher on Twitter's list.

If a topic is getting a Million tweets per hour steadily, and in that same period of time a different topic went form 5,000 tweets per hour to 50,000 tweets per hour, it would trend higher even though it's quantitatively being talked about much much less.

Now, the algorithm is more sophisticated than that so you can't game the system. It also factors in unique users, churn, and is heavily influencered on a Global scale by time zones and when territories wake up and go to sleep and whatnot. But in general, that's how it works.

Somebody tried shutting me down with the argument by sharing this image with me:
http://i.imgur.com/mFgx7xW.png

His claim was, "See, ParisAttacks should be trending higher than MuslimsAreNotTerrorists. Twitter is censoring!111!1!!"

The problem is, this person galvanized my point. Looking at the graph, while I see there are quantitatively more tweets about ParisAttacks than MuslimsAreNotTerrorists, there's actually a rate of DECAY. There's a deceleration happening. Meanwhile, MuslimsAreNotTerrorists accelerated during the same time period, and then held steady. PrayForWorld similarly was accelerating while ParisAttacks was decelerating.

This means the algorithm is working perfectly. Observing the Worldwide Trends, we see PrayForWorld above ParisAttacks (which is no longer on the list), and MuslimsAreNotTerrorists is there, too.

This is not Twitter censoring out the objective news to promote "progressive" ideas like Muslims not being Terrorists... Which, if you think that's a "progressive" idea, you might be clinically stupid because it should go without saying, regardless of a hashtag! It's a common sense idea, not progressive. But I digress. The algorithm is working as-expected.

One major digression happened when one person I was debating with didn't realize he was on the Tailored Trends view. He was creating all sorts of conspiracy theories around how A/B testing is happening, and how his Trends list is different than mine. So if your Trending Topics list doesn't say "Worldwide Trends" at the top, that means it's actually on Tailored Trends. If it just says "Trends" - that actually is a Tailored view, user-controlled.

To get Worldwide Trends back --

1. Click the "Change" link next to your Trending Topics list on the web.
2. Type "Worldwide" into the Search Locations box.
3. Press Enter.

Even Tailored Trends are based on a foundation of acceleration, mind you. And are generally filtered by region. So if I'm in Green Bay Wisconsin on Game Day, you bet I'll see #Packers as the #1 trend if I'm Tailoring by my location. That doesn't mean the Packers are more important than Paris.....

I hope that helps bring some insight into how Trending Topics works! Stay kind to one another, please :).

-Josh

Reply · Report Post