Did you catch the blog postblog post about the "Predict which questions will be closed" machine-learning contest?
The contest's closed now, and I don't know what the apparent winner's 0.29837 multiclass log loss score translates into in terms of accuracy, but robo-moderating is definitely under consideration.