We're scheduled to get some genuine spam dumps eventually, which is awesome. But I just figured out how to run Bayesian algorithms in Python, which just happens to be the language our friendly neighborhood Smoke Detector is written in. What a coincidence!
Yes, I'm planning on putting this learning into the spam-scanning chatbot. I think we can rely on users to supply new variants of spam through chat commands, and we can learn from those. The problem is getting the starting data.
It would be really awesome if we could be provided with a dump something similar to the following:
- Titles, and only titles, of questions destroyed as spam
- Over the whole network, if possible
- Over about the last month or two, or however much SE is willing to provide
Of course, this might come with some gah licensing problems, so it might be required to also add a link to to posting user's profile (even if they're destroyed). IANAL, though, and I don't really know what it would take.
That's really all I need to get started with the project. I think we can do some good to the world, we just need a place to start.