Learning to Rank Utterances for Query-Focused Meeting Summarization