I have a MediaWiki instance with thousands of articles in a certain scientific field. They're in a flat space with no categorization. I'd like to organize these automatically using data mining and language-processing techniques. In theory, I think looking for statistically unlikely phrases in each document would provide a good starting point.
Right now I can do something like that through the MediaWiki API -- pull down the documents, analyze them, and automatically write back categories or tags.
But is there another way to do this? Looking around the web shows that there's been a huge amount of work on this kind of problem in general -- but nothing that works specifically with MediaWiki in an automated, integrated solution. Is there such a thing?