One of the features of the new EveKnows.com is the ability to filter search results by sexual orientation. But with thousands of new galleries added every hour, how do we do this?
The answer is machine learning. Our search spider looks at the text of every gallery we index, along with the descriptions other sites use to link to those galleries. By starting with a small collection of galleries we know show “straight” content, and another collection of “gay” galleries, we can train a computer program to examine the gallery text and predict whether the pictures, tubes, or movies in the gallery are “straight” or “gay”. It’s the same approach Google uses to keep SPAM out of your Gmail inbox, and it works remarkably well. As some folks have noted, however, it’s not perfect. Yet.
The reason this technique is called “machine learning” rather than “machine prediction” is that our program continues to “learn” from new galleries that it indexes. So, over time, we’ll get better and better at distinguishing gay galleries from straight galleries. You might have noticed this over the past couple of weeks. Last week, our internal tests found about 50 straight picture galleries mistakenly classified as “gay” from a sample of 300 gay galleries. That’s about an 84% accuracy rate, which isn’t bad, but could definitely use some improvement. After continuing to learn all week, plus some algorithm changes we’ve made to help it along, today our spider misclassified only 6 galleries from another sample of 300. That means it was right 98% of the time!
With these sorts of improvements, we’re able to automatically index huge numbers of galleries every day, just like Google or Bing, without waiting for people to individually review each one. So, you get more free porn that’s accurately classified, helping you to find exactly what you’re looking for faster than ever before. It’s just one more of the ways we’re trying to make EveKnows become the world’s best porn search engine.
