Relation-aware Language-Graph Transformer for Question Answering (AAAI 2023)

Official pytorch implementation of 'Relation-aware Language-Graph Transformer for Question Answering' (AAAI 2023)

Setup

• Clone repository

git clone https://github.com/mlvlab/QAT.git
cd QAT

• Setup conda environment

conda create -n QAT python=3.8
conda activate QAT

• Install packages with a setup file

bash setup.sh

• Download data

We use the question answering datasets (CommonsenseQA, OpenBookQA, and MedQA-USMLE) and their knowledge graphs. We preprocess the dataset following QA-GNN. You can download all the preprocessed data with the link.

Run Experiments

bash run_csqa.sh

Acknowledgement

This repo is built upon the QA-GNN and GSC:

https://github.com/michiyasunaga/qagnn
https://github.com/kuan-wang/graph-soft-counter