Unverified50% confidenceFactExact time
Multi-label F1 computes precision and recall for each label separately and then takes the average (macro-F1) or averages over samples (micro-F1), but treats all errors alike and cannot reflect differences in semantic hierarchy.
1
Sources
50%
Confidence
Long-term
Relevance
11/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPrecision, recall, F1 score, and accuracy can all be derived from the four numbers of the confusion matrix.73% similarVerifiedThe F1 score is the harmonic mean of recall and precision.68% similarUnverifiedThe token-level F1 score separately computes the token-level precision and recall between the predicted sequence and the ground-truth sequence and then takes the harmonic mean, but it cannot distinguish semantically equivalent names with different spellings.68% similarUnverifiedUnder four-view test-time augmentation, ensembling the flat model with the label-correlation model achieved the best configuration on six of eight metrics: exact match 0.694, micro-F1 0.678, macro-F1 0.456, and macro-AUC 0.972.68% similarUnverifiedThe research team benchmarked five current mainstream LLMs, and on factual questions, the token F1 scores were only between 0.18 and 0.30 when no source passage was provided.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1182496API
curl https://kongchang.com/api/v1/knowledge/claims/1182496MCP
get_claim(id=1182496)