I've been tasked with coming up with a tool to automatically flag some posts for moderator attention on a large vbulletin forum. It's not spam per se, but the task has a lot in common with the sort of handling that might be done by a spam protection plugin (a mod in vbulletin speak). There's only so much I can say, but the task does not involve bad users, so much as particular kinds of posts which the moderators need to be aware of.
Filtering out user registrations and links is therefore not useful, and we are talking about posts by real human users. What I'm looking for is an existing bayesian classification plugin, or something that I can study to get an understanding of how to do the vbulletin side of the interface in order to build such a thing.
Ie I'd need ways for moderators to list flagged posts, and to correct the classification of posts which have been mis-classified. Ideally I want a 3 way split with an "unsure" category in order to reduce what has to be reviewed to find any mis-classifications.
Any pointers? I've searched around a bit, and so far what I've found has been more or less entirely targetted at intervening in sign-ups (mostly using stopforumspam), captchas, and use of external services like akismet which are spam specific.
I'm also considering an external solution, which might be ableto be interfaced i
