jianliu-ml / EEasMRC

=== Question Generation

1. Clone https://github.com/facebookresearch/XLM to the project.

2. Follow QuestionGeneration/data_preparation.py to prepare the training data. In the directory, train_source_example_inputs.txt gives samples in the source side, i.e., the narrative-style texts; train_target_what_questions.txt and train_target_when_questions.txt show samples in the target side, i.e., the query-style texts.

3. Train back-translation model to learn the mapping. generated_questions.txt gives the examples of the generated questions.

=== Question Answering

1. Follow util.py and model_train_progressly.py to preprocess the datasets and train the final model.