Japanese Financial SentiCSE
Sentiment-aware sentence embedding model for Japanese financial text, based on SentiCSE (Kim et al., COLING 2024).
Model Details
- Base model: Japanese RoBERTa (roberta-base)
- Training data: Japanese financial news (EWS dataset)
- Method: Sentiment-guided contrastive learning (SentiCSE)
- Best checkpoint: step 2500, val_macro_f1=0.7466, test_macro_f1=0.7366
Citation
If you use this model, please cite the original SentiCSE paper:
@inproceedings{kim2024senticse,
title={SentiCSE: A Sentiment-aware Contrastive Sentence Embedding Framework with Sentiment-guided Textual Similarity},
author={Kim, Jaemin and Na, Yohan and Kim, Kangmin and Lee, Sang Rak and Chae, Dong-Kyu},
booktitle={LREC-COLING 2024},
year={2024}
}