May 20, 2004

Findory

I had a great idea for a new website yesterday: personalized news by baysian spam filtering. It would work like this:


  • Collect RSS feeds for news sources and blogs
  • Set up individualized accounts
  • Present each user with a set of headlines. Based on which articles they click on, assign weights to the RSS description (perhaps with a Baysian spam filter). Allow users to indicate lack of interest, too.
  • Generate custom news feeds based on the results.

Talking about this at lunch today, I was informed an ex-Amazonian, Greg Linden, is working on exactly that. His site is called Findory. Apart from easily being able to express disinterest, it matches my vision closely.

I played with it for a few minutes today, reading about a dozen articles. Findory does an OK job of surfacing things that I am interested in, and I'm sure it will get better with usage. It is still shallow on content: a search for "monorail" turned up one article on the Las Vegas monorail. A search for "seattle monorail" turned up nothing, as did searches for "instapundit," "andrew sullivan" and "dan savage." My guess is that local news is generally thin, and blog results are non-existent. The former can probably be blamed on lack of RSS feeds, the latter on the narrow scope of the site. Findory is in beta-testing, so hopefully both will change in the near future.

Posted by awm at May 20, 2004 06:41 PM