YouTube And Shiva Rajaraman Bring Intelligent Channel Picks To Online Video

How do you create technology to recommend new channels to users? And when does a person need to step in to curate these recommendations? In this extended version of the conversation from our latest issue, we chat with Shiva Rajaraman, Group Product Manager for YouTube.

Shiva Rajaraman | Photo by Toby Burditt

"On YouTube, our top queries are 'funny videos,' 'LOL,' and 'make me laugh,'" YouTube group project manager Shiva Rajaraman told us recently. "They fall into this vague land of 'Entertain me!' So based on that, how do we help people discover fresh, fun content—without just sifting through the 48 hours of video that are uploaded every minute? We're still trying to figure that out fully, but we're starting by building out our channels. When users search for something broad, we want a more curated experience to appear, so 'funny videos' would bring up a landing page with humor clips from BuzzFeed, Fail Blog, Funny Or Die, and more. That way, they can fill a steady 15, 20, 30 minutes with us, instead of making new decisions every two minutes."

Below is the extended version of our recent Fast Talk.

How do you go about finding great content on a site as traffic-heavy as YouTube?

It's truly a problem of finding the diamond in the rough. We have 48 hours of video uploaded every minute. If you step back and try to put some context around that, that's the equivalent of having 200,000 full-length movies released a week. You've got so much information that mining it becomes a challenge. So we ask, "How are people finding those great videos today?" They blog about it, or they share it on a social network or email and they write a great headline for it, and that's often how you're discovering YouTube videos. Think back to things like the Hudson River plane landing. In our logs we saw a huge surge in searches for the official plane landing video, but also that every news site on the web was embedding it. It's important for us to be able to say, hey, someone's found a video that's attracted an audience. That's one way we kind of dig out a diamond and make sure we surface it across the site.

Could you give me an example?

There's a video I love called "Three Big Pigs," and it's this Angry-Birds-meets-Libya video. When I go look at "Three Big Pigs," it's clear I've got this nuanced sense of humor, probably like everyone else through Angry Birds, but I also like political humor. If I just looked at keywords for a video and took "three big pigs" and did a search on YouTube, I'll probably find a bunch of videos on three actual big pigs. Collaborative filtering allows you to look at the crowd who watched that video and look at what else they've watched. And we step back even further from that to look at a blog that blogged about that video and see what else they've blogged about. So what we're doing is using the crowd to think about what we might offer to someone else who watches a particular video.

And channels could link up all those related videos.

Imagine a world in which we have related channels. Right now you switch from video to video, but imagine a world in which you're switching from channel to channel, more like TV. That's the thing we need to move towards. Right now, we recommend channels on the home page for users on our site, not just individual videos, but you're going to see us do more there. We're on 300 million devices. On most of those devices, clicking and querying and hunting just isn't fun. When we launch on Xbox this year and I'm sitting in front of Xbox, I want to be able to use Kinect to say something like "Lego Animation" and have that channel start playing.

So what's more important to the curation process, man or machine?

It depends what you're into. When there's stuff that's not easy to bake into the video itself, I would like much more of an intro around that stuff. Rebecca Black is a good example of this. It's a really bad music video, but what's better than that is the story. I need someone to tell that story, and no machine can do that as effectively as a person today. That's where human curators come in. But sometimes, the content speaks for itself and I don't really care so much about the person who created it, I just want to see this awesome video. In that case, the machine is useful.

Read more Fast Talk

Add New Comment

1 Comments