Re: [Wikitech-l] Try #2: Idea for learning spam filter

The idea is that the normal behaviour for a wiki is that it is very unlikely for an url to be referenced in an edit more than X times in 24 hours (I'll postulate twice, just for fun), but it is very common behaviour for a spammer. Therefore, if one could ascertain (do some tests) that some large percentage (say 99.9%) of valid (non-spam) urls posted via edits occur less than X in 24 hours, you could put in a filter inconveniencing a very small percentage of users that would autmatically add those urls to a blacklist. I think it's worth testing anyway. What I'm proposing is to collect urls posted via edit in a short period of time (say a week) and do an analysis. Anybody else like this idea? I'd be willing to write a script, but there'd have to be a hook, or a small hack in editpage to call it.