Currently the spam score is derived from the text analysis of the comments and a couple of other account properties, not actually those in the bulleted list. I'm currently working on a better classification algorithm, as I think I should be able to get much better accuracy than I have at the moment.

I hope to provide more useful info soon too.