Pandora’s Podcast Genome Project Wants to Find Your Next Favorite Show


There are properly over 500,000 podcasts on the earth, protecting the whole lot from ABBA to zoology. They’ve reached such a saturation level, in actual fact, that there’s nearly actually a podcast on the market that’s completely attuned to your pursuits—and no nice method to discover it. Streaming radio veteran Pandora thinks it has an answer in—what else, in 2018—an algorithm. Its Podcast Genome Project, first introduced practically a 12 months in the past, launches Tuesday in beta.

- Advertisement -

That identify ought to sound acquainted; it was the Music Genome Project, in spite of everything, that propelled Pandora’s early success as a streaming music supplier. Its podcast-focused counterpart shares the identical underlying aims, slicing and dicing particular person episodes, utilizing greater than 1,500 tags to energy suggestions. It appears slightly totally different below the hood, although.

“In the case of podcasts, we will depend on machines extra closely than we will with music,” says Pandora CEO Roger Lynch. “Machines can really decide content material, decide intent, there’s many extra issues they will decide a couple of podcast than they will a couple of track.”

As with the Music Genome Project, people present each steerage and a backstop for algorithmic decisions. But they play a much less distinguished position with podcasts, no less than thus far. Instead, Pandora leans on pure language processing to parse the content material of a given podcast episode, assigning descriptors associated to content material sort, manufacturing fashion, the host profile, and plenty extra. If you’ve hit thumbs-up on a podcast episode that talks about, say, dad and mom and children rebuilding engines for traditional vehicles collectively, Pandora will discover you one other, says chief product officer Chris Phillips.

“You can think about that just about something that’s mentioned within the podcast is a candidate to change into a genomic trait,” Phillips says. As it occurs, the Podcast Genome Project already pulls from practically 4 occasions as many descriptors as its musical counterpart. “What we’re doing is beginning with what we will extract out that we all know is a subject somebody may search on, however we additionally know can develop, after which you may put mixtures of these along with context.”

Humans come into play as guardrails; pure language processing could not decide up on satire, as an illustration, or be capable to inform fictional tales from reality.

“Machines can really decide content material, decide intent, there’s many extra issues they will decide a couple of podcast than they will a couple of track.”

Pandora CEO Roger Lynch

The granularity of Pandora’s Podcast Genome additionally hints at a number of the challenges it would face. Phillips notes {that a} latest episode of the Questlove Supreme podcast tackled Atlanta politics at across the 50-minute mark, serving to Pandora floor it for anybody who has expressed curiosity in that pretty particular subject. At current, although, that listener must sit via practically an hour of unrelated dialog, or scrub till they discovered the related dialogue.

Which itself raises one other potential problem: Unlike songs, many podcasts are a considerable time dedication. You can solely hearken to so many in a day. Given that, how lengthy may it take for the Podcast Genome Project to garner sufficient data out of your listening habits to make really helpful recommendations?

Clearly it’s not not possible to create a suggestion engine for longer-form content material. Netflix and different video streamers do it on a regular basis. But whereas podcasts could also be simpler to categorize than music, they could show trickier to counsel.

“With a music suggestion, when you get one track mistaken or one track is off barely, it’s OK. Each track is simply two or three minutes. You can provide it a thumbs-up or thumbs-down, and fine-tune alongside the way in which,” says Erik Diehn, CEO of Stitcher, a podcast listening app that additionally owns Midroll Media, an promoting community that works with reveals like Freakonomics Radio and WTF with Marc Maron. “But while you’re making a selection a couple of podcast, the stakes are slightly increased. You have a smaller set of samples to select from as you attempt to determine what someone likes or doesn’t like.”

Pandora foresees a future during which it could possibly level folks not simply to particular podcast episodes, however related moments inside these episodes. In the extra speedy time period, Lynch says, the corporate is encouraging podcast producers to give attention to shorter codecs, bite-size morsels that characterize much less of a dedication, and clear the way in which for much less intrusive promoting.

Which, unsurprisingly, is the opposite half of Pandora’s podcast equation. Its music algorithms don’t simply assist listeners discover songs; additionally they assist advertisers discover the precise listeners. The firm sees much more potential in focused promoting for podcasts, which inherently cater to extra clearly outlined pursuits. And streaming a present via a platform may doubtlessly provide higher analytics, which in flip would assist podcast producers—and Pandora—cost increased charges.

“We have over 3,000 focused segments to promote advertisers,” says Lynch. “All the advertisements that we ship on Pandora are focused. If you examine that to podcasts, most podcasts are downloaded. There’s a measurement problem. Do you already know if the podcast was even listened to? And how lengthy was the podcast listened to? The reply to all of that is, you don’t know.”

At launch, most podcasts on Pandora will nonetheless have the host-read mattress advertisements to which you’re accustomed. But in time, Lynch hopes to switch lots of these with an audio type of the programmatic promoting that permeates a lot of the remainder of the web. The profit to the listener hinges on the ol’ “extremely related advertisements” argument—how significantly better do you actually need advertisers to know you, actually—that Facebook and others have made for years. But for podcast producers, particularly exterior of the headline acts, it may assist make their efforts extra sustainable.

“Most podcast gamers simply pull from an RSS feed. Monetization is left totally to the podcaster. That’s a great factor and a nasty factor, relying on the way you have a look at it,” says Diehn. “Obviously it makes it more durable for someone with no infrastructure, no gross sales group, no actual place to begin to generate significant income. There’s no YouTube for podcasts.”

Then once more, efforts to offer that form of advert scaffolding exist already, as do fashions for discovery, whether or not from an built-in podcast firm like Midroll, fashionable apps like NPR-owned Pocket Casts, or Spotify, or Google, which just lately launched its first native podcast app for Android. Unlike its music discovery engine, Pandora’s Podcast Genome Project received’t be the primary main participant to take the sector. And whereas the podcast challenge launches with content material from heavy hitters like Gimlet, American Public Media, The New York Times, and NPR, amongst others, it should nonetheless must aggressively construct out its steady to be able to present listeners with the really sudden.

At the very least, although, Pandora is giving podcasts a brand new potential path for achievement—and if its suggestion engine works as marketed, giving itself newfound relevance within the course of.


More Great WIRED Stories

- Advertisement -
Previous Xbox X018: Date, Time, Confirmed Games, and News
Next 'Spider-Man' PS4 Gets a New Patch In Time For 'Turf Wars' DLC