I'd be surprised if pg hasn't experimented with it at least a bit given his history with text classification algorithms.