MiniConGTS: A Near Ultimate Minimalist Contrastive Grid Tagging Scheme for Aspect Sentiment Triplet Extraction

Update

2025-03-30:

Fixed the missing files for your fast installing the reliances and getting started.

Now you can enjoy this repo by simply clone and run

conda env create -f environment.yml
python main.py

Have fun!

Overview

This repository serves as the official codebase for my recent work, "MiniConGTS: A Near Ultimate Minimalist Contrastive Grid Tagging Scheme for Aspect Sentiment Triplet Extraction". The project implements a minimalist tagging scheme and a novel token-level contrastive learning strategy to enhance aspect sentiment triplet extraction performance. The approach leverages the power of Pretrained Language Models (PLMs), such as BERT, RoBERTa etc. to achieve state-of-the-art results without relying on complex classification head designs or external semantic enhancements.

Arxiv Preprint

Key Features

• Minimalist Grid Tagging Scheme: Uses the fewest classes of labels to simplify the tagging process.
• Token-level Contrastive Learning: Improves the representation quality of PLMs, enhancing their internal potential.
• High Performance: Achieves state-of-the-art results in Aspect-Based Sentiment Analysis (ABSA) with minimal reliance on complex architectures.
• Evaluation on GPT Models: Includes first-time evaluations of the Chain-of-Thought method in context learning scenarios.
• Comprehensive Analysis: Provides proofs and theoretical analyses to support the efficacy of the proposed methods.

Repository Structure

MiniConGTS/
│
├── data/                   # Directory for datasets
│   ├── D1/
│   │   ├── res14/
│   │   │   ├── train.json
│   │   │   ├── dev.json
│   │   │   └── test.json
│   │   └── ...             # Other datasets
│   └── D2/
│       └── ...
│
├── modules/
│   ├── models/
│   │   ├── roberta.py      # RoBERTa model definition
│   │   └── saved_models/   # Directory for saving trained models
│   └── f_loss.py           # Focal loss implementation
│
├── tools/
│   └── trainer.py          # Training and evaluation script
│
├── utils/
│   ├── common_utils.py     # Utility functions
│   ├── data_utils.py       # Data loading and preprocessing
│   └── ...
│
├── main.py                 # Main script for training and evaluation
├── requirements.txt        # List of required packages
└── README.md               # This file

Getting Started

Installation

1. Clone the repository:

   git clone https://github.com/your-repo/MiniConGTS.git
   cd MiniConGTS

2. Install dependencies:

   conda env create -f environment.yml

Preparing the Data

1. Place your datasets in the ./data/ directory following the provided structure.

Training the Model

1. Configure training parameters in the main.py script or pass them as command-line arguments.

2. Run the training script:

   python main.py --max_sequence_len 100 --batch_size 16 --epochs 2000 --dataset res14

   for more parameter setting:

   python main.py \
    --max_sequence_len 100 \
    --sentiment2id "{'negative': 2, 'neutral': 3, 'positive': 4}" \
    --model_cache_dir "./modules/models/" \
    --model_name_or_path "roberta-base" \
    --batch_size 16 \
    --device "cuda" \
    --prefix "./data/" \
    --data_version "D1" \
    --dataset "res14" \
    --bert_feature_dim 768 \
    --epochs 2000 \
    --class_num 5 \
    --task "triplet" \
    --model_save_dir "./modules/models/saved_models/" \
    --log_path "./logs/training_log.log" \
    --learning_rate 1e-3 \
    --warmup_steps 500 \
    --weight_decay 0.01
   
   

Alternatively, you can start your jupyter kernal and debug each intermidiate step easily in a notebook using:

main.ipynb

Evaluation

The model is evaluated on the test set after training. The results, including metrics such as accuracy, precision, recall, and F1-score, are logged to a file specified in the log_path.

Citation

If you use this code in your research, please cite the paper as follows:

@inproceedings{sun-etal-2024-minicongts,
    title = "{M}ini{C}on{GTS}: A Near Ultimate Minimalist Contrastive Grid Tagging Scheme for Aspect Sentiment Triplet Extraction",
    author = "Sun, Qiao  and
      Yang, Liujia  and
      Ma, Minghao  and
      Ye, Nanyang  and
      Gu, Qinying",
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.emnlp-main.165/",
    doi = "10.18653/v1/2024.emnlp-main.165",
    pages = "2817--2834",
    abstract = "Aspect Sentiment Triplet Extraction (ASTE) aims to co-extract the sentiment triplets in a given corpus. Existing approaches within the pretraining-finetuning paradigm tend to either meticulously craft complex tagging schemes and classification heads, or incorporate external semantic augmentation to enhance performance. In this study, we, for the first time, re-evaluate the redundancy in tagging schemes and the internal enhancement in pretrained representations. We propose a method to improve and utilize pretrained representations by integrating a minimalist tagging scheme and a novel token-level contrastive learning strategy. The proposed approach demonstrates comparable or superior performance compared to state-of-the-art techniques while featuring a more compact design and reduced computational overhead. Additionally, we are the first to formally evaluate GPT-4`s performance in few-shot learning and Chain-of-Thought scenarios for this task. The results demonstrate that the pretraining-finetuning paradigm remains highly effective even in the era of large language models."
}

License

This project is licensed under the MIT License. See the LICENSE file for details.

Acknowledgments

Special thanks to the authors and contributors of the transformers library by Hugging Face.