Normalizing Citysearch
DeWitt Clinton noticed that Citysearch ratings have inflated in the past few years, and decided to do something about it. His greasemonkey script normalizes the ratings to pick out the true top. Check out his before and after pics.
This approach should be the front line filter for all semantic web trust problems. I envision a day when all metadata you browse can be normalized and filtered to strip out outliers, and match normal statistical models. I’ve also argued for years that you need to be able to cluster based on past activity patterns for you and your circle of friends. This is no different from what we do in real life — your brain sbconsiously clusters events into nameless contexts and gives weightings to observations based on context. The web should be the same.