Codes for the paper "Exploring Sequence-to-Sequence Learning for Aspect Term Extraction".
./code/train.py -> for training
./code/model.py -> for model details
./code/evaluation.py -> for testing
./code/A.jar -> offical script for restaurant domain
./code/eval.jar -> offical scirpt for laptop domain
./best_model/restaurant -> this is the best model trained by our model for restaurant domain
./best_model/laptop -> this is the best model trained by our model for laptop domain
./data -> store necessary data
OS : Ubuntu 16.04.4 LTS
Python : 3.6.8
Pytorch: 1.0.0
Numpy: 1.16.3
- We adopt the data processing method from the paper: 'Double Embeddings and CNN-based Sequence Labeling for Aspect Extraction' (https://arxiv.org/abs/1805.04601).
- If you want to process your own data, please follow their introduction (https://github.com/howardhsu/DE-CNN).
- We have preprocessed the dataset by their method, and all data are stored in dir: /data/pre_data/
CUDA_VISIBLE_DEVICES=0 python code/train.py laptop/restaurant
CUDA_VISIBLE_DEVICES=0 python code/evaluation.py laptop/restaurant
We must thank all authors from this paper: 'Double Embeddings and CNN-based Sequence Labeling for Aspect Extraction'. We adopt many codes from their projects. Thank a lot!
I apologize to all readers that I can not get the original results in the paper for some reason. I fine-tune on two datasets and get new results. It is unbelievable that the new results are higher than the results reported in the paper.
Restaurnat: 75.14 -> 76.15
Laptop : 80.31 -> 80.62
If you think the codes & paper are helpful, please cite this paper. Thank you!
@inproceedings{ma2019exploring,
title={Exploring Sequence-to-Sequence Learning in Aspect Term Extraction},
author={Ma, Dehong and Li, Sujian and Wu, Fangzhao and Xie, Xing and Wang, Houfeng},
booktitle={Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics},
pages={3538--3547},
year={2019}
}