Conversational Question Generation

Implementation for our ACL 2019 paper: Interconnected Question Generation with Coreference Alignment and Conversation Flow Modeling

Requirements

tqdm; pytorch=0.4.1; torchtext; numpy; python=3.6; cuda=9.0

Download Processed Data

Please clone this repo, download our processed data here and put it into data directory:

git clone https://github.com/Evan-Gao/conversaional-QG.git
cd conversational-QG
mkdir data

coqg-train/dev/test-3.json: our train/dev/test data split

coqg-coref-test-3.json: coreference test set

Provide the avaialble GPUs in a comma delimeted list following the bash script command, e.g. 0,1. You can find your available GPUs by issuing nvidia-smi

Preprocess

run scripts/preprocess.sh 0,1 for preprocessing.

GloVe vectors are required, please download glove.840B.300d first. run scripts/emb.sh for getting corresponding word embedding.

Train, Generate & Evaluate

run scripts/train.sh 0,1 for training, scripts/generate.sh 0,1 for generation and evaluation

Pretrained Model

We have released our pretrained model here.

Reference

If you use code, please cite our paper as follows:

@inproceedings{Gao2019InterconnectedQG,
	title="Interconnected Question Generation with Coreference Alignment and Conversation Flow Modeling",
	author="Yifan Gao and Piji Li and Irwin King and Michael R. Lyu",
	booktitle="ACL",
	year="2019"
}