Question Answering Corpus

This repository contains a script to generate question/answer pairs using CNN and Daily Mail articles downloaded from the Wayback Machine.

For a detailed description of this corpus please read: Teaching Machines to Read and Comprehend, Hermann et al., NIPS 2015. Please cite the paper if you use this corpus in your work.

Bibtex

@inproceedings{nips15_hermann,
author = {Karl Moritz Hermann and Tom\'a\v{s} Ko\v{c}isk\'y and Edward Grefenstette and Lasse Espeholt and Will Kay and Mustafa Suleyman and Phil Blunsom},
title = {Teaching Machines to Read and Comprehend},
url = {http://arxiv.org/abs/1506.03340},
booktitle = "Advances in Neural Information Processing Systems (NIPS)",
year = "2015",
}

Download Processed Version

In case the script does not work you can also download the processed data sets from [http://cs.nyu.edu/~kcho/DMQA/]. This should help in situations where the underlying data is not accessible (Wayback Machine partially down).

Running the Script

Prerequisites

Python 2.7, wget, libxml2, libxslt, python-dev and virtualenv. libxml2 must be version 2.9.1. You can install libxslt from here: http://xmlsoft.org/libxslt/downloads.html

sudo pip install virtualenv
sudo apt-get install python-dev

Download Script

mkdir rc-data
cd rc-data
wget https://github.com/deepmind/rc-data/raw/master/generate_questions.py

Download and Extract Metadata

wget https://storage.googleapis.com/deepmind-data/20150824/data.tar.gz -O - | tar -xz --strip-components=1

The news article metadata is ~1 GB.

Enter Virtual Environment and Install Packages

virtualenv venv
source venv/bin/activate
wget https://github.com/deepmind/rc-data/raw/master/requirements.txt
pip install -r requirements.txt

You may need to install libxml2 development packages to install lxml:

sudo apt-get install libxml2-dev libxslt-dev

Download URLs

python generate_questions.py --corpus=[cnn/dailymail] --mode=download

This will download news articles from the Wayback Machine. Some URLs may be unavailable. The script can be run again and will cache URLs that already have been downloaded. Generation of questions can run without all URLs downloaded successfully.

Generate Questions

python generate_questions.py --corpus=[cnn/dailymail] --mode=generate

Note, this will generate ~1,000,000 small files for the Daily Mail so an SSD is preferred.

Questions are stored in [cnn/dailymail]/questions/ in the following format:

[URL]

[Context]

[Question]

[Answer]

[Entity mapping]

Deactivate Virtual Environment

deactivate

Verifying Test Sets

wget https://github.com/deepmind/rc-data/raw/master/expected_[cnn/dailymail]_test.txt
comm -3 <(cat expected_[cnn/dailymail]_test.txt) <(ls [cnn/dailymail]/questions/test/)

The filenames of the questions in the first column are missing generated questions. No output means everything is downloaded and generated correctly.