Text classification based on pre-trained Bert model

[Cheng-Su]

1. Our goal in this Demo is to use the pre trained BERT model to complete the task of classifying items on e-commerce websites, and to judge whether the item belongs to the Fashion industry by entering the title of the item.
2. For the model, we selected HuggingFace Bert Base Cased pre training checkpoint.
3. The model training for an epoch has converged well. The following indicators are Accuracy, AUC, Precision, Recall, F1, etc. If we take Threshold=0.5, then

Accuracy	AUC	Precision	Recall	F1
0.9757	0.9943	0.9355	0.9389	0.9372