python 3.6.4
    tensorflow 1.8
    nltk 3.2.5
    keras 2.1.5
    sklearn 0.19
    gensim 1.7

    https://github.com/likejazz/Siamese-LSTM
    https://github.com/aditya1503/Siamese-LSTM
    https://blog.csdn.net/android_ruben/article/details/78427068

    the pre-trained word vector has been used(numberbatch-17-06_en in the test).both word2vec and numberbatch are recommended

    acc on training data: 0.9078
    acc on test data: 0.8316

may be improved by useing stemming of words and filter stopwords
to do:
    1.train new word embedding(numberbatch may not support stemming)
    2.process the input file into stemming version