trellixvulnteam/bert-hierarchical-softmax-chinese-text-classification_72re

Bert-Hierarchical-Softmax-Chinese-Text-Classification

bert+hierarchical softmax

pretrained model

Three files are needed in /bert_pretrain:

• pytorch_model.bin
• bert_config.json
• vocab.txt

All these files you can download here: bert_Chinese: model: https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese.tar.gz
vocabulary: https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-chinese-vocab.txt
reference from(https://github.com/huggingface/pytorch-transformers)

A backup link for pretrained model：https://pan.baidu.com/s/1qSAD5gwClq7xlgzl_4W3Pw

From the site you can get a zip file, unzip it and put it in /bert_pretrain.

Hierarchical softmax introduction：https://talbaumel.github.io/blog/softmax/

Data

Some samples are loaded:

./data/train.txt

where train.txt is a text file containing: title, content, label_level_1, label_level_2.

By default, the seperatotr is set to \t.

How to use

Just need to download pretrained files above.

# train：
# bert
python train_evl.py

Reference

[1] Hierarchical softmax：https://talbaumel.github.io/blog/softmax/

[2] BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

[3] https://github.com/649453932/Bert-Chinese-Text-Classification-Pytorch