README.md

This is a Pytorch implementation and released dataset of "FaiMA: Feature-aware In-context Learning for Multi-Domain Aspect-based Sentiment Analysis" accepted by LREC-COLING 2024.

Feature-aware In-context Learning for Multi-Domain Aspect-based Sentiment Analysis (FaiMA)

More details of the paper and dataset will be released after it is published.

The Code

Requirements

Following is the suggested way to install the dependencies:

pip install -r requirements.txt

Folder Structure

└── SA-LLM
    ├── data                    # Contains the datasets
    │   ├── inst/ASPE           # Our MD-ASPE instruction data
    │   ├── raw/ASPE            # MD-ASPE raw data
    ├── checkpoints             # Contains the trained checkpoint for model weights
    ├── src
    │   ├── gnnencoder          # The code related to MGATE
    │   ├── Icl                 # The code related to Feature-aware In-context Learning
    │   ├── llmtuner            # The code related to LLM train, predict etc.
    ├── run_gnn.py              # The code for training MGATE
    ├── run_aspe.py             # The code for training FaiMA and baselines
    └── README.md               # This document

Training and Evaluation

1. Run run_gnn.py to train MGATE model.
2. Run run_aspe.py to train FaiMA and baselines, replece model_name_or_path with your llama model weight path.