OTM recently looked at the phenomenon of "Reply Girls," the cleavage baring women crowding YouTube with nonsensical videos. YouTube says it is trying to fix the problem of irrelevant videos on its site. Bob speaks to YouTube engineering director Cristos Goodrow about how the site is changing its algorithm to show users more of what they want to see.
Smog - Held
BOB GARFIELD: Spending less time on the site is exactly what YouTube is afraid of. A few weeks after we ran that story, we heard from YouTube, which wanted to let us, and you, know about changes it’s making in its algorithm to make irrelevant videos less of an issue. And according to YouTube Engineering Director Cristos Goodrow, it isn’t just the “Reply Girls” causing the problem. CRISTOS GOODROW: We have seen instances in which videos will provide something misleading in the thumbnail or the metadata. So, for instance, there might be a video of someone talking about a fight, you know, a boxing match that happened, and the thumbnail has the boxing match where the guy’s just getting knocked out. But if you click on the video, all you see is the person talking about that match and there is no clip of the fight. BOB GARFIELD: So how does YouTube continue to welcome the kind of back and forth conversation among video makers but get rid of what essentially is video spam? CRISTOS GOODROW: One way we do it is to emphasize the videos that people are watching more than the ones that they’re just clicking on. And so, in particular, we recently changed the order of the recommendations we make to users to emphasize those that actually get watched over those that just get clicked. The change we made was not just in reaction to the Re – “Reply Girls” or view count gaming. It’s a broader change that actually improves the site across the board. It’s one that we’ve been working on since last year. We don’t really think of it so much in terms of fighting spam, but rather promoting good. BOB GARFIELD: So if I deem cleavage to be irresistible and click on a “Reply Girl” and find out that she’s just babbling nonsensically, and I abandon the video after only four seconds, and it’s actually a 2-minute video, you’ll pick that up and realize that the abandonment rate suggests lack of substance, entertainment, relevance. CRISTOS GOODROW: That’s right. It suggests lack of engagement with that video for the people who were watching the, the previous video. It’s not just about the percent of the video they watch. It’s about the amount of time they spend watching. We like to compare this to, say, shopping for a camera. You expect a salesperson to tell you which one most people purchase and use that to influence your choice.
Imagine instead if customers told the store how much they liked the camera after buying it. That would make for a better recommendation, and that’s what we’re trying to do with the videos on YouTube. BOB GARFIELD: One thing the “Reply Girls” did to game the algorithm was duplicate the tagging of the original popular video. What happens to a, a truly phenomenal video? Instead of just a hamster summersault, the hamster is also levitating and reciting The Gettysburg Address while in mid-air, right? [GOODROW LAUGHS] It’s a virtuosic rodent performance but is utterly untagged and simply posted. Does it slip through the cracks at YouTube? CRISTOS GOODROW: Well, we’re trying to close those cracks as quickly as we can. It does make it more difficult. We can use all the help we can get from the uploaders and also from the community in initially tagging the video, giving it a good title that’s meaningful and also watching it. When people watch it next to other videos, then we can gather that information too, to understand what the video’s about and also how popular it might be.
But it’s a hard problem that we’re working all the time to solve: How can we identify what videos are about from the information that we get on the video – from the tags, from the metadata, from the other sessions in which it gets watched – those things? BOB GARFIELD: Ultimately, even though Google has about a half-a trillion-dollar algorithm as its primary asset, in the end, can it defeat cleavage? CRISTOS GOODROW: [LAUGHS] Well, I don’t see it really as the algorithm defeating the cleavage. What’s actually happening here is the users themselves are defeating it. When the users don’t watch the whole video, then all we’re doing is keeping track of that information and using it to influence the next time we show that video. BOB GARFIELD: I was actually thinking of posting a video myself, just entirely bare-chested. [GOODROW LAUGHING] You think – do you think I’ll be enriched as a result? CRISTOS GOODROW: I think it depends on what you have to say while you’re bare-chested. BOB GARFIELD: [LAUGHS] Cristos, thank you very much. CRISTOS GOODROW: You’re welcome. [MUSIC UP AND UNDER] BOB GARFIELD: Cristos Goodrow is engineering director at YouTube.