Topics: Behind the Tweets

Thursday, 30 January 2020

Three months ago, we launched Topics to make it easier to find the most interesting conversations happening on Twitter about the things you love. 

Since then, people have asked to learn more about how Topics work. So, we want to provide some background on how specific Topics are selected, the process to find the best Tweets for you about these Topics, and how we seek to highlight healthy conversations. 

How Topics are chosen

Right now, there are roughly 1,000 Topics you can follow via the Topic browser, which will then show you Tweets about those Topics in your Home timeline. And we’re adding new Topics almost every week.

This post is unavailable
This post is unavailable.

We add new Topics based on the volume and health of conversations happening on Twitter. We start by looking at the volume of conversation on Twitter: how much people are Tweeting, Retweeting, replying, and liking Tweets about that Topic. The more conversation, the more likely it will be evaluated for inclusion in our next batch of available Topics. Topics reflect the broader, lasting conversations people have about the events, people, and things they discuss. Once we’ve identified a new Topic, we then shift our focus to identifying the most relevant Tweets. 

Finding the most relevant Tweets on a Topic

To find the full set of Tweets about a Topic, we use a combination of methods including looking at the text of Tweets as well as signals generated by machine learning solutions highlighting what Tweets might be about. At times, our teams and external partners may also help source accounts to be included in certain Topics.

To find the most relevant Tweets, we start by identifying the people on Twitter who are most interested in a given Topic using using a combination of machine learning, curation, and keywords.* This could mean they Tweet a lot about the Topic or interact a lot with Tweets about the Topic. From there, we find the Tweets that are most interesting to those people, using additional algorithms that look at many signals from those fan accounts including what types of content they engage with or specific accounts that are most relevant. Our systems check to confirm these Tweets are relevant to the Topic by looking at a series of related keywords.

Keeping Topics healthy

A healthy conversation about Topics is one where you want and feel safe to participate. We’ve implemented a number of protections so that the conversations included in a Topic should be healthy and free from abuse. This includes working to not recommend Tweets for a Topic if the engagement is manipulated or spammy. We look at a number of underlying signals, including how others interact with both the Tweet and the author — things like reports, blocks, and whether people have dismissed similar recommendations by clicking “show less often” or muting closely-related keywords. These are just some of the signals we use when ranking and filtering the group of Tweets until we have a set we are confident recommending for that Topic. And, to be clear: Tweets that criticize, satirize, or disagree within a Topic are a natural part of a healthy conversation and are eligible for inclusion. These protections are on top of all of the health protections we’ve already put in place for other parts of Twitter

In addition to the processes described above, there are quality assurance checks with a globally distributed team of people to confirm Topics are accurately highlighting relevant and healthy conversations. If Topics are unfollowed frequently or Tweets within Topics are reported often, these could be signals that those Topic Tweets might not be pertinent and healthy. If this is happening, we may pause the Topic temporarily. 

We know people are trusting us by following Topics and that Topics offer additional amplification to Tweets, so we want to make sure we’re amplifying relevant, healthy conversations. 

Personalizing the Topic for you 

When you follow a Topic, the Tweets you see about that Topic in your timeline are personalized. Your past activity — like your Tweets and Tweets you’ve liked, replied to, or Retweeted — are used to identify the Tweets you'll find most interesting. Also, if you engage more with one Topic, like Dogs, versus another Topic, like cats, we’ll show you more of the Dog Topic. (Sorry cat lovers!) 

To wrap it up, let’s use one of my favorite Topics, BLACKPINK, as an example. For the BLACKPINK Topic, accounts regularly Tweeting about and engaging with the KPOP band generally or a specific band member, like a fan account, are found. From there, Tweets that would be interesting to those fan accounts are gathered using the method described above. Throughout this process, our systems are at work ensuring we are surfacing the conversations you want to see.

From that narrowed pool, depending on what might be most interesting to you, say which band member you Tweet about or reply to Tweets about the most, you may see a Tweet about Lisa or Jennie.

After all that, the best BLACKPINK Topic Tweet for you is ready to be shown in your Home timeline!

This is just the start, we’ll keep working to improve Topics. We hope you like it. 👍

This post is unavailable
This post is unavailable.
This post is unavailable.

If you don’t see a Topic you want to follow yet, don’t worry. We’re adding new Topics almost every week. Topics are available in English, Japanese, Spanish, Portuguese, Arabic, and Korean. 

We’d love to hear how you are liking Topics. Do the Topics you follow show interesting Tweets? When you follow a Topic, do you see too many or too few related Tweets? We’ll keep tweaking our systems to make Topics better. You can give feedback on Topic Tweets by tapping the down arrow on the Tweet and choosing “Show less often.” 

*Updated July 2, 2020 to elaborate on how Tweets about a Topic are discovered and clarify that for certain Topics, our teams and external partners may help source accounts to be included.


This post is unavailable
This post is unavailable.