Classifying Documents with Large-texts Based on RNN&BERT

• Collected a Chinese corpus bigger than 5G, composing of news, poems(shi), Chinese ci, classical proses, papers
• Implemented machine learning algorithms including Naive Bayers and Logistic regression to classify
• Used word2vec, combined with TextCNN, TextRNN, TextRNN_Att, TextRCNN, DPCNN, Transformer with pytorch; compared the accuracies
• Implemented ERNIE, bert, bert_cnn, bert_dpcnn, bert_rnn, bert_rcnn with pytorch; compare the accuracies

Machine learning algorithms

RNN & CNN

• Models comparison

  Model	Accuracy
  TextCNN	0.8734
  TextRNN	0.8853
  TextRNN_Att	0.8859
  TextRCNN	0.8862
  DPCNN	0.8853
  Transformer	0.8752

• The best moder performance
  
• Confusion matrix
  

Bert & ERNIE

• Models comparison

  Model	Accuracy
  ERNIE	0.8869
  Bert	0.8890
  Bert_cnn	0.8819
  Bert_dpcnn	0.8777
  Bert_rnn	0.8809
  Bert_rcnn	0.8789

• The best moder performance
  
• Confusion matrix