Unverified50% confidenceFactExact time
在类别不平衡场景下F1分数比准确率更能反映模型对少数类的识别能力
1
Sources
50%
Confidence
Long-term
Relevance
8/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn class-imbalanced datasets, Macro F1 is extremely sensitive to rare class performance, where small threshold changes cause dramatic fluctuations in precision and recall for rare classes77% similarUnverified在类别不平衡场景下单纯的准确率可能具有误导性，需要结合混淆矩阵进行更细粒度评估69% similarUnverifiedtop-1%贪婪解码准确率只考察单token预测的一致性，无法捕捉量化对长序列生成质量的真实影响，微小偏差会在自回归生成中累积65% similarUnverified异构模型交叉审查因不同训练分布带来不同缺陷敏感性，能减少单一模型的同源偏见和认知盲区65% similarUnverified部分模型存在针对测试集过度优化的风险，导致榜单分数与真实场景表现出现偏差（Benchmark饱和）64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/821049API
curl https://kongchang.com/api/v1/knowledge/claims/821049MCP
get_claim(id=821049)