A company is using Amazon Bedrock's automatic model evaluation to assess a generative text summarization model they built. Which metric should they use to measure the model's accuracy?