The two similar sentences that need to be calculated are first spliced together, and then the overall coding information is obtained through the Bert model, and then the dimension is output through the fully connected layer to output the probability of similarity and dissimilarity. And this support both english and chinese sentence similarity calculation.
We are using BERT model from Google, so before you using it, you need to download the model.
Click here: https://storage.googleapis.com/bert_models/2018_11_03/chinese_L-12_H-768_A-12.zip
Decompress the zip file and put it into 'data/pretrained_model/'. A parameter file will be offered to you as well.
link：https://pan.baidu.com/s/1N9hg0yRCbE3IBd79OBCR6w
code：gu7t
After downloading, just put it in 'data/'.