Questions about the FAQ extraction approach and evaluation

[Jezemy]

Hi, this is a really great idea! I'd like to ask a few questions about the FAQ extraction approach:

• Completeness across multiple documents – How do you ensure the extracted FAQs are comprehensive enough, especially when the relevant knowledge is spread across multiple documents? How can we be confident that nothing important gets missed?
• Quality governance – How do you plan to handle issues like duplicate, conflicting, or outdated FAQs among the extracted results? Is there a strategy for ongoing maintenance?
• Evaluation & comparison – Have you conducted any evaluation comparing this approach with other methods, such as naive RAG, FAQ-based RAG, and knowledge graph-based RAG? I'd be curious to know how it performs in terms of effectiveness, efficiency, etc.

Thanks for your work, and I'm looking forward to hearing your thoughts!