EGSS

EGSS paper code

Prerequisites

• Python 3.6
• PyTorch == 1.4.0

Quick start:

• Dataset

All dataset under the directory of squad-processed, include [SQuAD corpus](bio, linguistic features,src, tgt). You need to run answer_to_entity.py,ans-entity_to_qtype.py,other_covert.py, and style_question.py in order to get the question type. Dependency parsing relations need to be obtained yourself using the Stanford dependency tools and converted to adj.txt via adj_json.py We utilize the glove embedding, please download the glove.840b.300d.txt and put it in the data.

• Data preprocess

Run following command:

1. python preprocess_data.py

Then we will get two files

1. data/embedding.pkl
2. data/word2idx.pkl

• Training

Run command:

python main.py

• Inference

Run command:

python main.py -train False

Then the output file will be save in directory of result/