Home

Awesome

Kaggle-Quora-Insincere-Questions-Classification

Kaggle新赛-基于BERT的fine-tuning方案baseline

submit.csv是测试集的预测结果

代码和项目基于BERT的中文tagging一样,仅提供关键fine-tuning代码和运行脚本

基于bert的验证集的结果:

classprecisionrecallf1-score
00.980.980.98
10.650.620.63
micro avg0.960.960.96
macro avg0.810.800.81
weighted avg0.960.960.96

基于tensor2tensor的验证集结果:

classprecisionrecallf1-score
00.980.960.96
10.230.190.21
micro avg0.920.920.92
macro avg0.590.570.58
weighted avg0.910.920.91