Incorporating prior knowledge into a transductive ranking algorithm for multi-document summarization