Why it matters
Relevance is fit, not proximity. A chunk about webhooks is irrelevant to a refund question even if both mention accounts. Irrelevant chunks crowd context and confuse citations.
How it works
Use rerankers, metadata boosts, and intent classifiers to demote tangential passages. Train reviewers to label relevance separately from correctness.
Example
Nintendo demotes API authentication chunks when users ask about billing refunds, even though both mention account settings. Relevance scoring prioritizes Refund Policy sections first.
Common mistakes
- 1Assuming embedding similarity equals business relevance
- 2Ignoring table-of-contents chunks that match every query weakly
- 3No intent routing for high-risk domains
Definitions are useful. Governed knowledge is better.
Wiki helps you turn these concepts into a real system for the knowledge behind your AI.
See Wiki in action