Web 3.0 : Look, Ma, No Keywords!

Three new Web services reinvent the way we look for music and images.

You won't search for media with keywords in the future-—you'll search for media with media. To find an image, you'll supply another image. To find a song, you'll supply another song. Don't believe it? Three new services—image-crunchers Like.com and Polar Rose, and music-matchmaker Pandora—have already taken the first steps toward this new breed of media search.


Web 3.0 : An Introduction

Web 3.0: Tim, Lucy, and The Semantic Web

Web 3.0 :The Other Semantic Web

Web 3.0 :Semantics and Search

Web 3.0 : A Web Beyond Words

Web 3.0 : Tomorrow's Web, Today

Web 3.0 : An Idiot's Guide to Web 3.0

Web 3.0 : Questions of Semantics

Web 3.0 : Look, Ma, No Keywords!

Web 3.0: Versions 4, 5, 6...



Three new Web services reinvent the way we look for music and images.

You won't search for media with keywords in the future-—you'll search for media with media. To find an image, you'll supply another image. To find a song, you'll supply another song. Don't believe it? Three new services—image-crunchers Like.com and Polar Rose, and music-matchmaker Pandora—have already taken the first steps toward this new breed of media search.

Today, when you search the Web for music and images, you're merely searching for the words that surround them. When you visit Google Image Search and type in "Steve Jobs," you aren't really looking for photos of Apple's CEO. You're looking for filenames and captions that carry those keywords—"Steve" and "Jobs"—hoping the right photos are somewhere nearby.

There's a sizable difference between the two. On any given image search, Google turns up countless photos completely unrelated to your query, even as it misses out on countless others that may be a perfect match. In the end, you're relying on Web publishers to annotate their images accurately, and that's a hit-or-miss proposition.

The situation is much the same with MP3s, podcasts, and other sound files. When trolling Web-based music services, you can run a search on "Elvis" or "Jailhouse Rock." But what if you're looking for music that sounds like Elvis? Wouldn't it be nice if you could use one song to find other similar songs?

Ojos and Polar Rose are tackling the image side of the problem. Last spring, Ojos unveiled a Web-based photo--sharing tool called Riya, which automatically tags your pictures using face recognition. Rather than manually adding "Mom" tags to all your photos of Mom, you can show Riya what she looks like, and it adds the tags for you. The service is surprisingly accurate, gaining a huge following from the moment it hit the Web, but Ojos quickly realized that the Riya face-rec engine—which also identifies objects and words—could be used for Web-wide image search.

That's a mammoth undertaking, but, with an alpha service called Like.com, the company is already offering a simple prototype. Today, Like.com is little more than a shopping engine. You select a photo of a product that best represents what you're looking for, and the service shows all sorts of similar products. But it's an excellent proof-of-concept.

Meanwhile, Polar Rose (www.polarrose.com) recently introduced a browser plug-in that does face recognition with any photo posted to any Web site. For the moment, it's just a means of tagging images automatically—much like Riya. But unlike Riya, it already works across the length and breadth of the Net.

The closest equivalent when it comes to audio is Pandora, from a group of "musicians and music-loving technologists" called the Music Genome Project. Since its inception in 2000, the group has analyzed songs from over 10,000 artists, carefully notating the music makeup of each track. Using this data and a list of your favorite artists, Pandora can instantly construct a new collection of songs that suit your tastes. Again, this is hardly a Web-wide search engine, and unlike the image services from Ojos and Polar Rose, it relies heavily on up-front human input. But it's a step in the right direction. True media search is closer than you think.