I have tried different clustering methods before, but did not receive enough satisfying results. Especially I had a look in what carrot can do, because I find the library pretty cool itself. That its built in Solr and can do everything on the fly was just a bonus. But the entities change too much, there is not enough similar data (each day) which result in poor clustering, the clusters are not meaningful - no matter what I did. Probably I did something wrong. After all, I wanted to look into that a little bit later again. And the main purpose was actually not provide a reader-App with hard-coded categories. The main purpose was a feed which is my personal "top news" based on what I "liked" before - having my own categories. Ping me on thomas@newscombinator dot com and I'll shoot you a beta login, maybe you'll find it useful.
The confusion of the name is already well noted, and a "re-branding" is not excluded. Considering its not even a real product, its weird to talk about a rebranding though :) - Paul forgive me the confusing naming. I had nagrgtr.com before (for News-Aggregator), but I found that domain not-so-good (=sucks) as well. I guess I will sit down again and brainstorm a little bit - suggestions welcome.
Thanks for the feedback.
Tom