for example, a code that would screen the content of posts and comments, and if the same series of words repeat in different posts and comments N times in a period T, then all revenue on these posts is put back in the pool. Of course more reflexion needs to be put into it, in order to consider all cases and not hurt genuine posts inadvertantly. A starting point would be to define the threshold for detection (N/T) as a function of the length of the post, of the flag ratio of the user, etc...
It is an interesting idea, and not ruling it out - but the code for both the Steem blockchain and steemit.com (condenser) is open source, so whatever rules are put in place, it would not be very hard for an abuser to figure out what they are and code around them.
This argument makes sense, however all is a questions of stats: How many spammers would have the knowledge necessary to do that?
Yet, on second thought, you just need one person to translate the code to english, and post it... So all algorythmic rules are out of the game...
On third thought maybe not, if there is an engagement button: what about an anonymous Spammy button? If too many 'spammies' appear on a user, there could be an alert for anti-spam / fishing moderators like @arcange. A spammer account is easy to recognise on inspection.
Communities will have something along these lines