KR-FinBert-SC
| Property | Value |
|---|---|
| Author | snunlp |
| Task Type | Text Classification, Sentiment Analysis |
| Language | Korean |
| Training Data Size | 13.22GB (6,379,315 lines) |
| Framework | PyTorch, Transformers |
What is KR-FinBert-SC?
KR-FinBert-SC is a specialized Korean language model designed for financial sentiment analysis. Built upon the foundation of KR-BERT-MEDIUM, this model has been further pre-trained on an extensive financial corpus and fine-tuned specifically for sentiment classification tasks. The model demonstrates superior performance with a 96.3% accuracy rate, outperforming other Korean language models in financial sentiment analysis.
Implementation Details
The model was trained on a comprehensive dataset combining corporate economic news from 72 media sources and analyst reports from 16 securities companies. The training process involved 5.5M steps with a maximum sequence length of 512, using a batch size of 32 and a learning rate of 5e-5. The training was conducted on NVIDIA TITAN XP hardware over approximately 67.48 hours.
- Dataset composition: 440,067 news titles with content and 11,237 analyst reports
- Total training data: 13.22GB across 6,379,315 lines
- Outperforms KR-BERT-MEDIUM, KcBert-large, KcBert-base, and KoBert in sentiment classification
Core Capabilities
- Accurate sentiment analysis of Korean financial texts
- Processing of news articles and financial reports
- Classification of positive and negative financial sentiments
- Domain-specific understanding of financial terminology
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its specialized training on Korean financial texts and its superior accuracy (96.3%) in sentiment classification, particularly in the financial domain. It combines general Korean language understanding with domain-specific financial expertise.
Q: What are the recommended use cases?
The model is ideal for analyzing Korean financial news, market reports, and corporate communications for sentiment analysis. It can be particularly useful for financial institutions, market analysts, and researchers working with Korean financial texts.
