I can't imagine the amount of channel noise generated by certain keywords. And for every one you'd have to do a time consuming search through the user's social graph(s).

There should be a function in the scraper that gives each user a unique ID, and retains an instance count of keyword 'hits'. Over time, the user would build up a 'heat rating' and IDs could be sorted by this. That way, people generating keyword noise over and over again could be separated from the general population for a closer look.