Unverified50% confidenceFactExact time
After unifying the top-k sweep from the author's previous post onto the same 9 questions, the curve flattened between 0.67 and 0.83, with all seven runs falling within the judge noise band, so the sweep could not distinguish among the k values; the earlier conclusion of 'optimal k=8' does not hold.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedTop-K retrieval refers to returning the top K chunks with the highest similarity during the retrieval stage, and K is usually taken from 3 to 10.66% similarUnverifiedAfter fixing TOP_K to 8, the top score is 0.635, the second row is 0.592, and the language/education chunk at the end of the list scores below 0.45, with a visible separation reappearing at the top.66% similarUnverifiedThe value of K in Top-K should be determined through offline evaluation by measuring Recall@K at different K values on an annotated test set and finding the inflection point where the recall curve begins to flatten.64% similarUnverifiedUnder pure vector retrieval with k=5, only 2 of 12 questions could assemble all the necessary evidence; traversing along the graph reached 11/12.62% similarUnverifiedK折交叉验证的K通常取5或10，这在偏差-方差权衡上被实证为较优选择61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065193API
curl https://kongchang.com/api/v1/knowledge/claims/1065193MCP
get_claim(id=1065193)