README.md

Sentic GCN: A Simple But Effective Framework for Aspect-Based Sentiment Analysis via Affective Knowledge Enhanced GCN

Introduction

This repository was used in our paper:

Aspect-Based Sentiment Analysis via Affective Knowledge Enhanced Graph Convolutional Networks
Bin Liang, Hang Su, Lin Gui, Erik Cambria, Ruifeng Xu. Knowledge-Based Systems, 2021: 107643.

Please cite our paper and kindly give a star for this repository if you use this code.

Requirements

• Python 3.6
• PyTorch 1.0.0
• SpaCy 2.0.18
• numpy 1.15.4

Usage

• Please download Glove Embedding, and put 'glove.42B.300d.txt' under the main directory.

• Install SpaCy package and language models with

pip install spacy

and

python -m spacy download en

• Generate dependency graph with

python generate_dependency_graph.py

• Generate sentic graph with

python generate_sentic_graph.py

• Generate sentic & dependeny graph with

python generate_sentic_dependency_graph.py

Training

• Train with command, optional arguments could be found in train.py & train_bert.py

• Please tune the argument of --seed for better performance

• Run senticgcn: ./run_senticgcn.sh

• Run senticgcn_bert: ./run_senticgcn_bert.sh

Testing

• Testing with the models saved in state_dict. Optional arguments could be found in infer.py & infer_for_bert.py
• Please run python infer.py for the testing of non-BERT models.
• Please run python infer_for_bert.py for the testing of BERT-based models.

Citation

The BibTex of the citation is as follow:

@article{liang2021aspect,
  title={Aspect-based sentiment analysis via affective knowledge enhanced graph convolutional networks},
  author={Liang, Bin and Su, Hang and Gui, Lin and Cambria, Erik and Xu, Ruifeng},
  journal={Knowledge-Based Systems},
  pages={107643},
  year={2021},
  publisher={Elsevier}
}

See Also

• The original knowledge base of SenticNet could be found at https://sentic.net/downloads/.
• The knowledge source used in this code is SenticNet 5, which is stored at senticnet-5.0/senticnet5.txt.
• We also set several variants of our model:

  Model	Decription
  affectivegcn.py	Only using the affective information (i.e. Eq.2) for building graphs
  attsenticgcn.py	Combining our model with attention mechanism
  sdgcn.py	Interactively performing the graph convolutional operation based on dependency (i.e. Eq. 1) and affective (i.e. Eq.2) graphs

Credits

• The affective knowledge used in this work is from SenticNet.
• Here, we would like to express our heartfelt thanks to all the authors of SenticNet.
• The code of this repository partly relies on ASGCN & ABSA-PyTorch.
• Here, we would like to express our gratitude to the authors of the ASGCN & ABSA-PyTorch repositories.