Combining clustering and co-training to enhance text classification using unlabelled data