It's many pronged approach. User timelines are loaded for "reputable" users, (see blog for computing reputation) and search input is used with primative bayesian filtering.
In my experience, I have found Tweet source (web, Tweetdeck, API etc.) to be one of the most effective quality/authenticity indicators.
Using the tweet source is such a smart idea! So far I rely on username filtering / patterns. Toughest is to get rid of job postings ! Good luck with Tagwalk ! You've got a new user ;)
There is tons of job tweets, but most have "job" in the username and/or app source. Also any app source that contains "bot" or "feed" is most likely safe to ignore.
It already works for any hashtags or usernames that it encounters, here's #sundaybaconclub http://tagwalk.com/tag/sundaybaconclub