Sort:  

I just thought of something. How about taking into account if they have a valid proof photo as a metric? That would mean looking into their posts for something tagged introduceyourself, then parsing it to look for a photo. It gets a bit hairy while looking for the steemit written in paper, but I guess that's the general gist. I know not everyone has a verification post (I certainly don't) but it could help weeding out bots.