
..

🗨️ TikTok is great because you must 'dislike'

Source: Reply All

With TikTok you can only view one piece of content at a time, and it’s very clear when you completed viewing the entire piece of content: “did you get to the end of the video?”. In order to view the next video, you must swipe away. By swiping away before the end of the video, you indicate dislike or at least disinterest. A very clear “not interested” signal plus extreme computer vision to determine what is in a video gives the TikTok algorithm huge amounts of detailed data on what you will watch and what you won’t. Text or image based social media cannot rely on you scrolling away to know if you didn’t like it. They can only know if you reacted to the post or not.

Granted, TikTok technically also only knows if it engaged you or not. It can’t differentiate between watching and hate watching. But the podcast argues that that’s still a fuller dataset than typical social media sites.

Transcript from source

ANNA: So, according to the experts I talked to when somebody uploads a video to TikTok, that video is going, it’s going under a microscope where the algorithm is looking at it and scanning for what’s in the video.

Some experts, including Eugene, call them attributes. It’s things like, is there a dog at 32 seconds in? You know, is somebody showing their face? What is the person saying? What are the subtitles saying?

EUGENE: But it’s not as simple as that because every video can have dozens and dozens of features.

ANNA: Yeah.

EUGENE: What music is used, what’s the setting? Uh, what are the tags that are applied to that video?

ANNA: All of these attributes, it seems like they come together to make a long dictionary definition of what exactly is in the video.

EMMANUEL: Mmm.

ANNA: The next thing that 100 percent know for sure that it’s doing is, it’s monitoring your attention. And like, what that means is that it’s watching how far you’re making it into a video. It’s watching to see if you comment on something, if you share it. Um, all of those actions are signs that you’re enjoying the video and that you want to keep seeing more like it.

…

ANNA: So, that’s the basics of how the TikTok recommendation algorithm seems to work. And like, we interact with algorithms like this all the time. Like, everywhere you go, you’re going to run into an algorithm that’s trying to recommend you stuff. So for Netflix, it’s movies. YouTube, it’s just like, shorter videos. But like, the thing is, all of those recommendations feel pretty bad to me, honestly? I never watch anything that Netflix recommends me. But I enjoy most everything that TikTok serves me. It’s just so much better at taking that next step that those other algorithms aren’t doing.

And so, I wondered why? Like, what is it doing differently?

And after talking to Eugene and the other experts, I think it boils down to two things that TikTok is just doing better than anybody else.

The first thing: it’s getting a really clear idea of what exactly we like and don’t like. And the way it’s doing that has everything to do with how the app itself is designed. Like I said, TikTok is an incredibly simple app. It just shows you one video at a time.

EUGENE: That is pretty unique.

ANNA: Yeah.

EUGENE: And the only other app that, uh, I think has something comparable is an app like Tinder.

ANNA: I was not expecting you to say Tinder. I was like, what is it? [LAUGHTER] Tinder, oh, my God. It, it kinda is, yeah.

EUGENE: Right? Yeah. Tinder created the iconic swipe right/swipe left—

ANNA: Uh-huh.

EUGENE: …um, user interface interaction. And that’s important in that case because they, they do want to match you to people that you have clearly, explicitly said you’re interested in.

ANNA: Yeah.

EUGENE: And the same was the, uh, case for TikTok.

[MUSIC]

ANNA: TikTok is designed so that, if you are enjoying a video, you just keep watching it. You do nothing. And then, as soon as you see something in that video that you don’t like, or something that even like, mildly disinterests you, you swipe away. So TikTok figures out very quickly what you’re not interested in. And Eugene says that is really, really rare.

EUGENE: A lot of our social media today is only positive sentiment oriented.

ANNA: Mm-hmm.

EUGENE: There’s, there’s no dislike button on Facebook.

ANNA: Yeah.

EUGENE: Uh, there’s no dislike button, necessarily, on Twitter. And when you only capture positive sentiment—

ANNA: Uh-huh.

EUGENE: … the danger is you have a blind spot to things that mildly annoy or disturb people.

ANNA: Uh-huh.

EUGENE: In real life, humans are very attuned to this. You know, if you’re with your friends or your family or your significant other, and you do something that bothers them, they might not actively come out and say, “Oh, you’re annoying me,” or something like that. But you pick up on their body language—

ANNA: [LAUGHS] Yeah. Yeah.

EUGENE: … and you realize, you, you know, and you adjust, uh, based on that. That’s a really important feedback loop in just the social world generally.

ANNA: Yeah.

EUGENE: And TikTok figured out an interface that allowed them to capture positive and negative sentiment really cleanly in these short videos.

ANNA: Yeah. Like body language.

EUGENE: Yeah, it almost is a form of that.

ANNA: So, the second reason that, like, TikTok is just like, blowing everything else in, like, the recommendation world, like, out of the water—

EMMANUEL: Mm-hmm.

ANNA: …is just like, the amount of information that it’s collecting about the users. So, like, just to think about Netflix, you’re going to Netflix dot com. You’re picking one movie, and you’re watching it for two hours.

And so, compare that to TikTok. Those videos are short. They’re like, one minute long. So, you’re giving TikTok feedback on, like, 40 or 50 videos in the same amount of time that you would spend watching a movie. And what that means is you’re just shooting a fire hose of data into TikTok’s algorithm.

And Eugene says it’s that fire hose that has pushed TikTok’s algorithm into just, like, a league of its own.

Don't want to do that? @ me on twitter or mastodon

Every post on this blog is a work in progress. Phrasing may be less than ideal, ideas may not yet be fully thought through. Thank you for watching me grow.