representation-based method
基本的文本匹配模型,用于重复问题检测。模型使用词语和字符两种级别的嵌入向量,将两段文本进行LSTM表示,然后拼接得到隐含向量,进行二分类。
features
- pretrained word and char embedding
- combine word-level and char-level matching signal
- clone the reposity and run
sh scripts/setup.sh
cd data/atec
and download data from - atec data- split the origin csv file to
train.csv
,dev.csv
andtest.csv
- run
sh scripts/train.sh