This repository is the official implementation of the SIGIR 2022 Paper Personalized Abstractive Opinion Tagging.
If you have any question, please open an issue or contact keninazhao@163.com.
We provide a runnable version of the PATag data. You can run this program directly using data.pkl and matrix.pkl in Google Drive. You should download and move it under./DataSet/.
You can view the details of the relevant data through ./DataSet/data_info.txt, and ./DataSet/sample_for_data_pair.txt is an instance of data pair.
If you want to get raw data and more processing details, please use this Google form to submit your information and request access to PATag.
- CUDA >= 10.0
- Python >= 3.6
- PyTorch >= 1.7Check the packages needed or simply run the command:
pip install -r requirements.txtYou can run the program with the following command:
bash script/run.sh model_name gpus is_train'model_name' can be 'POT', 'POT_woBehavior' or 'POT_woHGAT', to correspond to our proposed model, as well as two variants.
We support and recommend using multiple GPUs for training and a single GPU for testing.
bash script/run.sh POT 0,1,2,3 1For reproducibility purposes, we place the model checkpoints at Google Drive. You should download and move it under ./Output/model_name/, then you can run the trained models to test by using best.pkl and best_memory.p.
bash script/run.sh POT 0 0If you find our code useful, please cite our paper as follows:
@article{zhao2022personalized,
title={Personalized Abstractive Opinion Tagging},
author={Zhao, Mengxue and Yang, Yang and Li, Miao and Wang, Jingang and Wu, Wei and Ren, Pengjie and de Rijke, Maarten and Ren, Zhaochun},
year={2022}
}