My top item these days, however, is actually bot or troll farm detection.
Social media is cool because you get to participate and overhear conversations that you normally would never have anything to do with. However, bot abuse on website like Twitter and Facebook means that every interaction sooner or later gets brigaded by bots amplifying their message.
The technology exists, and as far as I understand, it isn't particularly complex. Projects like BotSentinel do a surprisingly good job of identifying these bad actors.
Perhaps all this falls under the "Customizable moderation" header, but part of me thinks this would be best addressed in combination of some sort of "proof of work", be it as simple (and inaccessible) as a $5 fee on sign-up or something more accessible but more complicated.