CHIME: Cross-passage Hierarchical Memory Network for Generative Review Question Answering

This repository contains PyTorch implementation of the corresponding COLING 2020 Paper.

Breif Introduction

CHIME is a cross-passage hierarchical memory network for generative question answering (QA). It extends XLNet introducing an auxiliary memory module consisting of two components: the context memory collecting cross-passage evidences, and the answer memory working as a buffer continually refining the generated answers. A sample of syntactically well-formed answers show the efficacy of CHIME.

Dependency

python 3.7.7
apex 0.1
bert-score 0.3.4
BLEURT 0.0.1
nltk 3.4.5
rouge 1.0.0
torch 1.4.0
torchtext 0.5.0
transformers 2.8.0

Data

AmazonQA: Raw data can be downloaded from AmazonQA Project
Details of building experimental subsets are included in build_data.py. The subsets we used are uploaded here

Running

Change task type for model training, testing, predicting, analyzing and evaluating. In particular, analyzing refers to reveal model's interoperability by listing intermediate answers

python3 run_chime.py --devices 0,1 --model xlnet-base-cased --root_path YourDataPath --data_size 1.0 --epochs 3
--batch_size_perGPU 1 --model_output_path YourModelPath --prediction_output_path YourPredictionPath
--evaluation_output_path YourEvaluationPath --rev_num 10 --ans_num 1 --task Train

Licence

MIT

Citation

@inproceedings{chime2020,
    author = {Lu, Junru and Pergola, Gabriele and Gui, Lin and Li, Binyang and He, Yulan},
    title = {Chime: Cross-passage Hierarchical Memory Network for Generative Review Question Answering},
    booktitle = {The 28th International Conference on Computational Linguistics},
    year = {2020}
}

LuJunru / CHIME