HLG.pytoch

Pytorch implementation of "Enhancing Chinese Pre-trained Language Model via Heterogeneous Linguistics Graph", ACL 2022

Usage

All configuration are described in args.py:

python3 run_class.py --help
python3 run_match.py --help

Preparatory

Please download pre-trained language models and configure the corresponding path in args.py.

Train

python3 run_class.py --exp_name <exp name> --do_train

About Dataset

Part of the processed datasets are uploaded in dataset folder. If you want to conduct experiments on other datasets, please follow the format of these datasets.

About Graph Construction

Please refer to collate function in dataloader/classifier.py or dataloader/matcher.py.

About Model and the Multi-Step Information Propagation

Please refer to model.py.

About Trained Models/Checkpoints

Not released, will consider if necessary.