Skip to content

goodbai-nlp/RGAT-ABSA

Repository files navigation

Investigating Typed Syntactic Dependencies for Targeted Sentiment Classification Using Graph Attention Neural Network

This repository contains the code for the paper "Investigating Typed Syntactic Dependencies for Targeted Sentiment Classification Using Graph Attention Neural Network", IEEE/ACM Transactions on Audio, Speech, and Language Processing (TASLP)

Setup

This code runs Python 3.6 with the following libraries:

  • Pytorch 1.2.0
  • Transformers 2.9.1
  • GTX 1080 Ti

You can also create an virtual environments with conda by run

conda env create -f requirements.yaml

Get start

  1. Prepare data

    • Restaurants, Laptop, Tweets and MAMS dataset. (We provide the parsed data at directory dataset)

    • Downloading Glove embeddings (available at here), then run

      awk '{print $1}' glove.840B.300d.txt > glove_words.txt
      

      to get glove_words.txt.

  2. Build vocabulary

    bash build_vocab.sh
    
  3. Training

    Go to Corresponding directory and run scripts:

    bash run-MAMS-glove.sh
    bash run-MAMS-BERT.sh
    
  4. The saved model and training logs will be stored at directory saved_models

  5. Evaluating trained models (optional)

    bash eval.sh path/to/check_point path/to/dataset
    bash eval-BERT.sh path/to/check_point path/to/dataset
    

Results

GloVe-based Model

Setting Acc F1 Log Pretrained model
Res14 83.55 75.99 log model
Laptop 78.02 74.00 log model
Tweets 75.37 74.15 log model
MAMS 82.02 80.99 log model

BERT-based Model

Setting Acc F1 Log Pretrained model
Res14 86.68 80.92 log model
Laptop 82.34 78.94 log model
Tweets 76.28 75.41 log model
MAMS 84.52 83.74 log model

References

@ARTICLE{bai21syntax,  
	author={Xuefeng Bai and Pengbo Liu and Yue Zhang},  
	journal={IEEE/ACM Transactions on Audio, Speech, and Language Processing},   
	title={Investigating Typed Syntactic Dependencies for Targeted Sentiment Classification Using Graph Attention Neural Network},   
	year={2021},  
	volume={29}, 
	pages={503-514},  
	doi={10.1109/TASLP.2020.3042009}
}

About

Code for our paper "Investigating Typed Syntactic Dependencies for Targeted Sentiment Classification Using Graph Attention Neural Network" in TASLP

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published