The use of unlabeled data to improve supervised learning for text summarization