Unverified50% confidenceBenchmarkExact time
After removing the three leakage sources—metadata, news agency labels, and duplicate documents—F1 drops only from 0.9935 to 0.9814, a decrease of 1.21 percentage points.
1
Sources
50%
Confidence
Long-term
Relevance
9/23/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在评估阶段对候选关系标签进行随机打乱后，该模型的F1从37%骤降到14%71% similarUnverifiedCSP improves the F1 score for rule consistency determination from 0.90 to 0.94.67% similarUnverifiedFine-tuned DistilBERT achieves F1=0.9993 on in-distribution testing, but loses 12.9 mean average precision points under topic shift, a greater degradation than the linear model's 5.2 points.66% similarUnverifiedFrom epoch 160 to epoch 290, the MSE loss decreased from 0.0690 to 0.0652, a drop of only approximately 0.006 over 130 epochs.66% similarUnverifiedAfter removing the candidate filtering component, FEVER accuracy dropped to 76.24%.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/942804API
curl https://kongchang.com/api/v1/knowledge/claims/942804MCP
get_claim(id=942804)