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:
class | precision | recall | f1-score |
---|---|---|---|
0 | 0.98 | 0.98 | 0.98 |
1 | 0.65 | 0.62 | 0.63 |
micro avg | 0.96 | 0.96 | 0.96 |
macro avg | 0.81 | 0.80 | 0.81 |
weighted avg | 0.96 | 0.96 | 0.96 |
基于tensor2tensor的验证集结果:
class | precision | recall | f1-score |
---|---|---|---|
0 | 0.98 | 0.96 | 0.96 |
1 | 0.23 | 0.19 | 0.21 |
micro avg | 0.92 | 0.92 | 0.92 |
macro avg | 0.59 | 0.57 | 0.58 |
weighted avg | 0.91 | 0.92 | 0.91 |