Improve content classifier with better logic/rules

[Firstyear]

Currently this is based on some simple-naive file rules, but we should probably improve these rules to be smarter about how we look at the urls given. Potentially regexs or similar matching techniques could be used.