Filtering user-generated content

A Bear Sterns analyst's report on Google caused a stir early this month by estimating that its YouTube subsidiary made $15 million in all of 2006. One reason for the comparatively paltry sales was YouTube's reluctance to sell ads next to videos with uncertain provenance -- the company had to be sure the clip wasn't violating copyrights before it was monetized. Meanwhile, the entertainment industry was pressing YouTube, other user-generated content sites and file-sharing networks to police themselves, rather than forcing copyright holders to identify and complain about specific files.

These overlapping needs have created a powerful demand for software that recognizes files as they're uploaded onto user-generated sites or swapped on file-sharing networks. Not surprisingly, the tech industry has responded. The Motion Picture Association -- the global version of the MPAA -- is in the final stages of studying video-recognition technologies supplied by 11 firms and a university.

Included are a couple of usual suspects -- Audible Magic, whose customers include the iMesh file-sharing network, and Gracenote -- as well as such globally flavored upstarts as Advestigo, which was founded by a pair of French researchers, and Vobile, whose founders are Chinese engineers.

The basic approach is to assemble a database of unique identifiers for movies, TV shows, music videos and other forms of content, then compare files against this database as they're transmitted online. Within a couple of weeks, the MPA is expected to give several of those technologies an informal seal of approval.

At that point, look for more user-generated sites and file-sharing networks (e.g., StreamCast's Morpheus) to plug in content-recognition technologies and start experimenting with a variety of ways to sell advertising. According to executives at Advestigo and inside Hollywood, the rationale for identifying files has changed dramatically in the past year or so. The major studios used to view video "fingerprinting" techniques as a way to block copyrighted materially from being shared online. Now they see them as the key to building advertiser-supported business models -- they enable content owners to find out who their online audience is, as well as monitoring how ads flow to those viewers.

Much work remains to be done to complete the infrastructure -- for example, studios and content-recognition firms have to build their databases of unique identifiers. But when the MPA's report comes out, it could be the online equivalent of a gun going off in the race to port TV's business model to the Web.

MPA Content Recognition RFI Study Participant Companies

1. Advestigo
2. Audible Magic
3. Auditude
4. Gracenote
5. Intellivision
6. Magix/AudioID
7. NTT
8. Philips
9. Thomson
10. University of St Andrew, Scotland
11. VidyaTel
12. Vobile

http://opinion.latimes.com/bitplayer/2007/03/filtering_userg.html