Document Level Multi-Aspect Rating Prediction
This repository is a pytorch implementation for the following arxiv paper:
Deliberate Self-Attention Network with Uncertainty Estimation for Multi-Aspect Review Rating Prediction
Tian Shi, Ping Wang, Chandan K. Reddy
- Python 3.6.9
- argparse=1.1
- torch=1.4.0
- sklearn=0.22.2.post1
- numpy=1.18.2
Some data sources:
- TripAdvisor-R: https://github.com/HKUST-KnowComp/DMSC
- TripAdvisor-RU: https://github.com/Junjieli0704/HUARN
- TripAdvisor-B: https://github.com/HKUST-KnowComp/VWS-DMSC
- BeerAdvocate-R: https://github.com/HKUST-KnowComp/DMSC
- BeerAdvocate-B: https://github.com/HKUST-KnowComp/VWS-DMSC
Please download processed dataset from here. Place them along side with DMSC_FEDA.
|--- DMSC_FEDA
|--- Data
| |--- trip_rate
| |--- trip_binary
| | |--- dev
| | |--- dev.bert
| | |--- glove_42B_300d.npy
| | |--- test
| | |--- test.bert
| | |--- train
| | |--- train.bert
| | |--- vocab
| | |--- vocab_glove_42B_300d
|--- nats_results (results, automatically build)Training, Validate, Testing python3 run.py --task train
Testing only python3 run.py --task test
Evaluation python3 run.py --task evaluate
keywords Extraction python3 run.py --task keywords
Attention Weight Visualization python3 run.py --task visualization
If you want to run baselines, you may need un-comment the corresponding line in run.py.
| Model | BRIEF |
|---|---|
| mtCNN | CNN + Multi-Task Learning |
| mtRNN | Bi-LSTM + Multi-Task Learning |
| mtAttn | mtRNN + Self-Attention |
| mtBertAttn | BERT + Multi-Task Learning + Self-Attention |
| mtAttnDA | mtRNN + Deliberated Self-Attention |
| MtAttnFE | mtAttn + Pretrained Embedding + Feature Enrichment |
| FEDA | mtAttnDA + Pretrained Embedding + Feature Enrichment |
Coming Soon.
@article{shi2020deliberate,
title={Deliberate Self-Attention Network with Uncertainty Estimation for Multi-Aspect Review Rating Prediction},
author={Shi, Tian and Wang, Ping and Reddy, Chandan K},
journal={arXiv preprint arXiv:2009.09112},
year={2020}
}