Multi-Feature and Multi-Channel GCNs for Aspect Based Sentiment Analysis

This repository contains the code for the paper "Multi-Feature and Multi-Channel GCNs for Aspect Based Sentiment Analysis".
Please cite our paper and kindly give a star for this repository if you use this code.

Setup

This code runs Python 3.7.0 with the following libraries:

• Pytorch 1.13.1+cu116
• Transformers 2.9.1
• spacy 2.0.18

Get start

1. Prepare data

   • Restaurants, Laptop, Tweets and MAMS 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. Build aspect-graph and inter-graph

   • Go to the common folder:

   • Generate aspect-focused graph with

   python focused_graph.py
   

   • Generate inter-aspect graph with

   python inter_graph.py
   

4. Training

   Go to Corresponding directory and run scripts:

   bash run-MAMS-glove.sh
   bash run-MAMS-BERT.sh
   

5. The saved model and training logs will be stored at directory saved_models

6. 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
   

7. Notice Please remove the comments in the code to adapt it to different datasets.

Credits

The code of this repository partly relies on InterGCN & RGAT & DM-GCN.
I would like to extend my appreciation to the authors of the InterGCN, RGAT, and DMGCN repositories for their valuable contributions.