This repository provides the implement of the model described in the paper BLCU_NLP at SemEval-2018 Task 12: An Ensemble Model for Argument Reasoning Based on Hierarchical Attention.

The experiment is based on TensorFlow platform.

The model_code directory contains the train/dev/test data and codes of hierarchical attention model. To get the predict answer, run

    $ cd model_code
    $ python model_tf.py

The answer will be saved as test_answer in the current directory.

The ensemble directory contains scripts for majority voting. Put the different dev answer files into dev_answers directory, and run

    $ perl voting_dev.pl 

The accuracy after voted will be print on the screen, and the new answer file will be saved as dev_voted_answer

Put the different test answer files into dev_answers directory, and run

    $ perl voting_test.pl 

The accuracy after voted will be print on the screen, and the new answer file will be saved as test_voted_answer

The results of our experiments are as shown.