README.md

---
library_name: transformers
tags:
- sentence-classification
- korean
- multi-class
- ko-sroberta
- transformers
---

# Model Card for Sentence Type Classification

This model is fine-tuned to classify Korean financial sentences into four categories: Predictive, Inferential, Factual, and Conversational. It is built upon `jhgan/ko-sroberta-multitask`, a multilingual transformer model specialized for Korean NLP tasks.

## Model Details

### Model Description

- **Developed by:** Kwon Cho
- **Shared by:** kwoncho
- **Model type:** RoBERTa-based transformer (fine-tuned for sequence classification)
- **Language(s):** Korean (한국어)
- **License:** Apache 2.0 (from base model)
- **Finetuned from model:** [`jhgan/ko-sroberta-multitask`](https://huggingface.co/jhgan/ko-sroberta-multitask)

This model was fine-tuned for multi-class classification using supervised learning with Hugging Face Transformers and PyTorch.

### Model Sources

- **Repository:** [More Information Needed]
- **Demo:** [More Information Needed]

## Uses

### Direct Use

The model can be used to classify financial sentences (in Korean) into one of the following categories:
- **Predictive** (예측형)
- **Inferential** (추론형)
- **Factual** (사실형)
- **Conversational** (대화형)

### Training Data

- **Dataset Name:** 문장 유형(추론, 예측 등) 판단 데이터  
- **출처:** [AIHub 링크](https://www.aihub.or.kr/aihubdata/data/view.do?pageIndex=1&currMenu=115&topMenu=100&srchOptnCnd=OPTNCND001&searchKeyword=예측형&srchDetailCnd=DETAILCND001&srchOrder=ORDER001&srchPagePer=20&srchDataRealmCode=REALM002&aihubDataSe=data&dataSetSn=71486)

이 데이터는 한국어 금융 문장을 다음 네 가지 유형으로 분류합니다:
- `예측형 (Predictive)`
- `추론형 (Inferential)`
- `사실형 (Factual)`
- `대화형 (Conversational)`

### Out-of-Scope Use

- Not suitable for general-purpose Korean sentence classification outside financial or economic contexts.
- May not perform well on informal or highly colloquial text.

## Bias, Risks, and Limitations

- The model may carry biases present in the training dataset.
- Misclassifications could have downstream implications if used for investment recommendations or financial analysis without verification.

### Recommendations

Use this model in conjunction with human oversight, especially for high-stakes or production-level applications.

## How to Get Started with the Model

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("kwoncho/sentence_type_classification")
model = AutoModelForSequenceClassification.from_pretrained("kwoncho/sentence_type_classification")

text = "해당 종목은 단기적으로 하락할 가능성이 있습니다."
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
outputs = model(**inputs)