此篇章將講解QACNN github coding

File structure

Implementation of Query-based Attention CNN for Text Similarity Map (https://arxiv.org/abs/1709.05036)

The file structure include 6 folders and 1 main script (main.py) should be included :

preprocess/
    plot2vec.py
    qa2vec.py
word_vec/
    register.json
    glove.42B.300d.json
raw_data/
    plot/
        IMDB_KEY.split.wiki
    question/
        qa.json
output_data/
    plot/
    question/
utility/
    utility.py
model/
    MODEL.py
main.py

usage

1. put the data given by MovieQA into raw_data folder (including IMDB_KEY.split.wiki and qa.json)
2. download GloVe word2vec and put it into word_vec folder (GloVe: https://drive.google.com/file/d/0B3UsyrHYzsTzekMtbndMZXB1Ync/view?usp=sharing )
3. run 2 scripts (plot2vec.py & qa2vec.py), and all the preprocessed data will automatically be saved to output_data folder
4. run main.py

Reference

[0] QACNN: Query-based Attention CNN https://github.com/chun5212021202/QACNN