Using corpus and knowledge-based similarity measure in Maximum Marginal Relevance for meeting summarization