metterian / numnet

NumNet: Machine Reading Comprehension with Numerical Reasoning

This is the implementation of NumNet: Machine Reading Comprehension with Numerical Reasoning. The code is based on NAQANet.

Introduction

Numerical reasoning, such as addition, subtraction, sorting and counting is a critical skill in human's reading comprehension, which has not been well considered in existing machine reading comprehension (MRC) systems. To address this issue, we propose a numerical MRC model named as NumNet, which utilizes a numerically-aware graph neural network to consider the comparing information and performs numerical reasoning over numbers in the question and passage.

Prerequisite

To use this source code, you need Python3.6+, a few python3 packages and DROP data. The python dependencies can be installed as follows:

pip install -r requirements.txt

Data

• augmented_drop_dataset_train.json: The augmented training set.

• comparison.json: This is the “Comparison" set mentioned in the paper. The samples are built by manually swapping answer candidates in the question of the corresponding examples.

Usage

Before training and inference, please download DROP dataset and put it in the data directory

Training

To train the model, you can use the following command. Here [MODEL_PATH] is the directory where you want to save your model.

allennlp train ./config/config_for_train.json -s [MODEL_PATH] --include-package numnet

Inference

To perform inference, the following command can be used. Here, [MODEL_PATH] is the directory where you save your model, and [FILE_TO_PREDICT] is the data you want to do inference on.

python predict.py  --include-package numnet --archive_file [MODEL_PATH]  --input_file [FILE_TO_PREDICT]  --output_file ./predictions.json

Citation

@inproceedings{ran2019numnet,
    title={{N}um{N}et: Machine Reading Comprehension with Numerical Reasoning},
    author={Ran, Qiu and Lin, Yankai  and Li, Peng  and Zhou, Jie  and Liu, Zhiyuan},
    booktitle={Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)},
    pages={2474--2484},
    year={2019}
}