In this repository, we provide checkpoints of the meta-trained models developed using our SLIM framework. SLIM models are optimized for sentiment classification and rating classification tasks in few-shot settings. You can download the models via Hugging Face or Google Drive.

Use the following code to load one of the models listed below.

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("hjkim811/SLIM-W_polar")
model = AutoModelForSequenceClassification.from_pretrained("hjkim811/SLIM-W_polar")

• SLIM-C_boolean: https://huggingface.co/hjkim811/SLIM-C_boolean
• SLIM-C_mixed: https://huggingface.co/hjkim811/SLIM-C_mixed
• SLIM-C_polar: https://huggingface.co/hjkim811/SLIM-C_polar
• SLIM-W_boolean: https://huggingface.co/hjkim811/SLIM-W_boolean
• SLIM-W_mixed: https://huggingface.co/hjkim811/SLIM-W_mixed
• SLIM-W_polar: https://huggingface.co/hjkim811/SLIM-W_polar

• All models: https://bit.ly/slim_models

• SLIM-W_polar is the best-performing model and is recommended for use.
• For more details on the framework and models, please refer to our paper.

If you find this repository useful, please cite our work:

@article{kim2025sentiment,
  title={Sentiment lexicon integrated meta-training for sentiment analysis in data-scarce settings},
  author={Kim, Hyunjong and Cho, Sungzoon},
  journal={Journal of Intelligent Information Systems},
  pages={1--30},
  year={2025},
  publisher={Springer}
}