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)

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

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
   

@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}
}