TechCrunch’s Erick Schonfeld has an interview with myYearbook CEO Geoff Cook. myYearbook is a social network that launched live video chat in November. With some mileage behind the feature, Cook is ready to share some data and results.

At the time the feature launched, 1 of every 10 streams on Chatroulette was “obscene,” he says. Since then, however, Chatroulette has made progress. Recently, myYearbook reviewed 1,500 Chatroulette streams and found that there was an abuse rate of 1.9% (or around 1 in 50). Cook claims a rate of “less than 1 in 1000” for myYearbook, which serves up 750,000 video chat conversations a day.

They are able to accomplish this through a combination of human and technological elements. The service captures and analyzes thousands of images each second. These images come from streams that are chosen at random.

From an algorithmic perspective, they have found that an image with a face in it is 5 times less likely to also contain nudity. Of course, they don’t rely on that alone as they have also found that 20% of images with a face also contain nudity. Cook says that “with a lot of effort and additional processing logic including many factors like chat reputation, social graph, motion, etc.,” they are able to bring it all together and more accurately label images that are “safe.”

In order to mitigate false positives, they also send a sampling of “safe” images for human review. “Safe” images are sampled at a lower rate than images that are not marked “safe.”

Their human review system works in two stages: essentially, if two reviewers decide that your actions are inappropriate, you are banned from the site. Their system allows the reviewers to check 400 images a minute and both tiers of their reviewer system are employed 24 hours a day, 7 days a week, 365 days a year.

Based on our findings, we believe purely algorithmic approaches to moderation will never provide adequate safety,” Cook explains.

One other interesting note: Cook believes that providing a login, “any login,” is a more valuable deterrent for abuse than forcing identity through Facebook Connect or something similar because “so long as there is any login, a user’s device can be blocked.” They use Threatmetrix to fingerprint devices that are used to access their service and when someone is banned, both the login and the device are banned.

Because of their success moderating live video, they plan to roll out a similar system for the review of photos uploaded to myYearbook.