In the absence of persistent identity (it's unclear whether this is a feature of your application or not), this is basically an impossible task. With persistent

In the absence of persistent identity (it's unclear whether this is a feature of your application or not), this is basically an impossible task. With persistent identity, there are a few options. If you can afford invite-only growth, then tracking invite trees like on the other site works remarkably well. When you find a spammer, you can nuke an entire subtree of accounts, because it is rare for bad-faith users to invite good faith ones, and the opposite (good faith users inviting bad faith ones) is rare enough that it cuts down on the number of bad faith users substantially anyways. Another "codification of trust"-type system is having comments by new authors hidden to most users by default, except to a random sample trusted users, who can choose to vouch for the message or actively reject it. Users with abnormally bad taste in vouches/rejections can be removed from the pool of users who randomly get assigned approval duty. I used a web forum that operated like this back in the day.