This project by Italian artists Andy Simionato and Antonio Riello aims at serving only the good news. In contrast to other good news sites, the news is not manually selected but by machine. So I had an opportunity to use some nice AI features for machine learning and language processing.
I have made a spamfilter-like system that uses a naive bayesian classification algorithm which has to be trained to correctly categorize news items into good and bad items.
The news items are gathered from popular news sites by automatically aggregating rss feeds every hour. For this I used SimplePie, which works great.
After aggregation, the items are parsed and put through the system for categorization.
I also used the Porter stemmer algorithm for reducing words to their stems to facilitate the automated learning process.
Furthermore, I built an easy to use backend in which the filters can be trained and managed. And of course I did the frontend as well.
Visit the site at http://www.onlythegood.org.