practicingman/chinese_text_cnn

TextCNN Pytorch实现 中文文本分类

论文

Convolutional Neural Networks for Sentence Classification

参考

• https://github.com/yoonkim/CNN_sentence
• https://github.com/dennybritz/cnn-text-classification-tf
• https://github.com/Shawn1993/cnn-text-classification-pytorch

依赖项

• python3.5
• pytorch==1.0.0
• torchtext==0.3.1
• jieba==0.39

词向量

https://github.com/Embedding/Chinese-Word-Vectors
（这里用的是Zhihu_QA 知乎问答训练出来的word Word2vec）

用法

python3 main.py -h

训练

python3 main.py

准确率

• CNN-rand 随机初始化Embedding

    python main.py
  

    Batch[1800] - loss: 0.009499  acc: 100.0000%(128/128)
    Evaluation - loss: 0.000026  acc: 94.0000%(6616/7000)
    early stop by 1000 steps, acc: 94.0000%
  

• CNN-static 使用预训练的静态词向量

    python main.py -static=true
  

    Batch[1900] - loss: 0.011894  acc: 100.0000%(128/128)
    Evaluation - loss: 0.000018  acc: 95.0000%(6679/7000)
    early stop by 1000 steps, acc: 95.0000%
  

• CNN-non-static 微调预训练的词向量

    python main.py -static=true -non-static=true
  

    Batch[1500] - loss: 0.008823  acc: 99.0000%(127/128))
    Evaluation - loss: 0.000016  acc: 96.0000%(6729/7000)
    early stop by 1000 steps, acc: 96.0000%
  

• CNN-multichannel 微调加静态

    python main.py -static=true -non-static=true -multichannel=true
  

    Batch[1500] - loss: 0.023020  acc: 98.0000%(126/128))
    Evaluation - loss: 0.000016  acc: 96.0000%(6744/7000)
    early stop by 1000 steps, acc: 96.0000%