Hacker News Reader

Comment by paulsingh | Hacker News Reader

Building a noise filter for Twitter: http://www.philtro.com

Very very interesting.

If it's not supposed to be secret, how does it work? Does it look for similarity of what you like or does it weight people in your feed and just skip x% of tweets from heavy low-value tweeters?

My instant thought is it simply trains a naive Bayes or similar spam classification algorithm on the tweets you label ;)

Actually, that's exactly what I did in the early stages of the project. Once I started getting some traction earlier this year, it made sense to start putting some real effort into building a much better engine.

Today, that's still one of the methods we employ but it's certainly not enough to make solid recommendations or wrap IP around... so, I've built a team of people much smarter than me to help take this thing to the next level. :)

Well, most of the "secret sauce" is wrapped into the algorithms that we've built so I can't say too much. In general, we're able to learn what topics you're interested in via data that we collect through your implicit and explicit actions. (i.e., we're learning about you by watching who you tend to interact with, retweet, what topics you tend to click through on, who you follow, etc.)

We've been building and improving the algorithm for a few months now -- it's certainly got a long way to go still but, on the other hand, it's certainly come along way so far. :)

FYI for others, there's also http://filttr.com

Reply on news.ycombinator.com