Content filter CrowdSifter cleans up with $50K from former Sex.com owner

Someone is finally mobilizing the collective wisdom of the porn-viewing masses. Well, sort of. CrowdSifter, the new content filter out of startup Dolores Labs, gathers and shows batches of images to a pool of thousands of public viewers who, together, flag obscene and violent images within minutes. The product sharpened its teeth earlier this week with an influx of more than $50,000, from Match.com founder and former Sex.com owner Gary Kremen (he sold the latter for an impressive $12 million in 2006).

To use the service, you sign up on the CrowdSifter website. It then crawls your own site for images, uploads them in a public forum where they are rated immediately, and provides you a report of results. As you can imagine, this could come in handy particularly for sites where users post their own profiles and photo galleries.

Content moderation (especially when it comes to pornography and user-generated content), is a pretty crowded field, as Dolores chief executive Lukas Biewald well knows. But CrowdSifter is one of the only players outsourcing flagging duties to a broader Internet public (reached through Amazon Mechanical Turk, an online marketplace connecting people with jobs computers just aren’t that good at yet). These viewers receive varying micropayments for their work.

This puts a unique spin on the concept that saves time, and especially money, Biewald says. Names like Neocom Multimedia and eModeration have established filtering businesses, but most other players outsource their services to retained contractors. Not only are these resources more expensive, but they actually require more time than the near-instantaneous combined effort of a large crowd.

That being said, there are a couple sites that pose more direct competition — namely TagCow, a site that also utilizes Mechanical Turk to tag and filter photos (amateur moderators are paid about 3 cents per tagged photo). While their website states that the service can be used to detect potential obscenity, its prime focus is to tag objects in photos to make them more easily searchable through major engines. Because its only one prong of TagCow’s business, CrowdSifter may be able to edge them out on that count.

It will also gain and steep advantage if it successfully rolls out similar services for audio, video and text content (all of which are being tested, and could be expedited by the recent financing). All of these tasks are still easier to accomplish with human resources than automated programs, though Biewald said he wouldn’t be surprised if computers become better at detecting the nuances of pornography in the next several years. More difficult will be finely-tuning these filters to different levels. After all, content on a site targeted at young children may require stricter scrutiny than photos on MySpace, a distinction that humans are still better at gauging.

In the press release, Kremen said he chose to invest in the service because he believes that Internet content should be properly targeted to appropriate user bases. Biewald first got the idea for CrowdSifter after he launched FaceStat, a site similar to HotorNot.com, and was struck by the number of obscene images users were posting. He was already familiar with Mechanical Turk at the time, and saw the opportunity for not only instantaneous moderation, but multiple reviews of each image by moderators, allowing for greater accuracy.