Metrics such as ROUGE and F1 can be inaccurate because they rely on shallow linguistic similarities (word overlap) between ground truth and LLM responses, even when the actual meaning differs.

Authors

• Person: Not available Organization: Amazon Web Services
  Evaluating RAG applications with Amazon Bedrock knowledge base ...

Sources

• Evaluating RAG applications with Amazon Bedrock knowledge base ... aws.amazon.com Amazon Web Services via serper

Referenced by nodes (3)

• ROUGE concept
• ground truth concept
• F1 concept