Question 35
Domain 3: Knowledge Integration, Data Handling, Cognition, Planning, and MemoryWhat technique improves retrieval precision?
Correct answer: A
Explanation
Reranking improves precision by taking a broad initial set and then scoring it with a stronger model. As stated, "retrieve top-100 with fast first-stage retrieval, rerank with cross-encoder model to get best top-10," which filters out weaker matches and leaves the most relevant results for generation.
Why each option is right or wrong
A. Implement reranking: retrieve top-100 with fast first-stage retrieval, rerank with cross-encoder model to get best top-10, use reranked results for generation.
Precision is improved by a two-stage retrieval pipeline: a high-recall first pass returns a broad candidate set (for example, the top 100), and a second-pass cross-encoder rescoring step selects only the strongest matches (for example, the top 10). In information retrieval terms, this reduces false positives by applying a more expensive relevance model to the candidate pool before generation, which is exactly the mechanism that increases precision rather than recall.
B. Retrieve only top-3 documents.
C. Use better embedding model.
D. Increase chunk overlap.