There is this pretty cool research project at Stanford (it's headed by Andrew Ng of the Coursera fame) that uses NLP and machine learning to detect content that the user might prefer not to see. It's still pretty rough around the edges but this demo is pretty cool.
http://www.youtube.com/watch?v=dQw4w9WgXcQ