This repository contains source code for paper "An End-to-End Deep Framework for Answer Triggeringwith a Novel Group-Level Objective" is accpeted by EMNLP 2017. (paper)
We use the WikiQA data set. Please see the paper for more details: WikiQA: A Challenge Dataset for Open-Domain Question Answering
Please download the original WikiQA code package, and run the following commands for preprocessing:
cd WikiQACodePackage/code
python -u process_data.py --w2v_fname ../data/GoogleNews-vectors-negative300.bin --extract_feat 1 ../data/wiki/WikiQASent-trai n.txt ../data/wiki/WikiQASent-dev.txt ../data/wiki/WikiQASent-test.txt ../wiki_cnn.pkl
Our model use the exact same preprocessed data here: "../wiki_cnn.pkl" for fair comparison of performances.
We use tensorflow (0.12.1) to implement our NN model. Copy "../wiki_cnn.pkl" into "./data" before running the following command:
python run.py --train --plus_cnt=True
python run.py --test --plus_cnt=True
Please kindly cite our paper if you use the code in this repo:
@inproceedings{zhao2017end,
title={An end-to-end deep framework for answer triggering with a novel group-level objective},
author={Zhao, Jie and Su, Yu and Guan, Ziyu and Sun, Huan},
booktitle={Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing},
pages={1276--1282},
year={2017}
}