DAGCN: Dual-channel and Aspect-aware Graph Convolutional Network for Aspect-based Sentiment Analysis in Computational Social Systems

This repository contains the code for the paper "DAGCN: Dual-channel and Aspect-aware Graph Convolutional Network for Aspect-based Sentiment Analysis in Computational Social Systems", IEEE Transactions on Computational Social Systems.

Requirements

numpy==1.23.5
pandas==2.0.3
torch==2.1.2+cuda
transformers==4.42.3

To install requirements, run pip install -r requirements.txt.

Preparation

Prepare data
- Restaurants, Laptop, Twitter and MAMS datasets. (We provide the parsed data at directory dataset)
- Downloading Glove embeddings (glove.840B.300d.zip), and put it into glove directory after unzipping the file.
- Downloading pretrained BERT (bert-base-uncased) and put it into bert-base-uncased folder.
Build vocabulary
```
bash build_vocab.sh
```

Training

Go to Corresponding directory and run scripts:

sh run.sh

The saved model and training logs will be stored at directory results automatically.

Results

GloVe-based Model

Database	Acc	F1	Log
Res14	84.90	78.34	log
Laptop	79.59	76.60	log
Twitter	76.96	75.92	log
MAMS	81.96	81.09	log

BERT-based Model

Database	Acc	F1	Log
Res14	87.67	81.56	log
Laptop	81.49	78.45	log
Twitter	78.14	77.61	log
MAMS	85.03	84.63	log

References

If you find this work useful, please cite as following.

@ARTICLE{10588956,
  author={Wanneng Shu and Cao Zhai and Ke Yu},
  journal={IEEE Transactions on Computational Social Systems}, 
  title={DAGCN: Dual-Channel and Aspect-Aware Graph Convolutional Network for Aspect-Based Sentiment Analysis in Computational Social Systems}, 
  year={2024},
  doi={10.1109/TCSS.2024.3418472}}

Credits

The code and datasets in this repository are based on DualGCN_ABSA and R-GAT, and we thank them.

zhaicao/DAGCN-ABSA