hongyuntw / RBRS-ARG

Code for ARG

This repository is built on top of Neu-Review-Rec, and includes various functionalities such as data preprocessing, modeling, and training RBRSs, among others.

Training RBRSs

To train RBRSs, please follow the training flow of Neu-Review-Rec and run the following command:

python3 main.py train \
--dataset ${dataset} \
--model=${model}

Training ARG

Training LM on Reviews Corpus

To train the language model on the reviews corpus, please run:

python3 lm/train_lm.py

Training Aspect Extractor (ABAE)

The implemented ABAE is based on this repository. To train/test the aspect extractor, run:

python3 aspect/train_abae.py \
--dataset ${dataset} 

python3 aspect/test_abae.py \
--dataset ${dataset} 

Leave-One-Out Pretraining on Transformers-Based Encoder-Decoder

To perform leave-one-out pretraining, you can obtain the data from Leave one out data can obtained from SelSum and run:

python3 leave-one-out/train.py

Train ARG

Finally, to train the ARG, please run:

python3 main.py train_arg.py \
--dataset ${dataset} \
--model=${model} \
--pth_path="checkpoints/${model}_${dataset}_default.pth" #RBRS checkpoints