YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

Korean Claim Detection Model for Fact-Checking

๋ชจ๋ธ ์†Œ๊ฐœ (Model Description)

์ด ๋ชจ๋ธ์€ ํ•œ๊ตญ์–ด ๋ฌธ์žฅ์—์„œ ํŒฉํŠธ์ฒดํฌ๊ฐ€ ํ•„์š”ํ•œ ์ฃผ์žฅ(claim)์„ ์ž๋™์œผ๋กœ ํƒ์ง€ํ•˜๋Š” ์ด์ง„ ๋ถ„๋ฅ˜ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

๋‰ด์Šค ๊ธฐ์‚ฌ, ์ •์น˜ ํ† ๋ก , ์†Œ์…œ ๋ฏธ๋””์–ด ๊ฒŒ์‹œ๋ฌผ ๋“ฑ์—์„œ ๊ฒ€์ฆ์ด ํ•„์š”ํ•œ ์ฃผ์žฅ์„ ์‹๋ณ„ํ•˜์—ฌ, ํŒฉํŠธ์ฒดํ‚น ์›Œํฌํ”Œ๋กœ์šฐ์˜ ์ฒซ ๋‹จ๊ณ„๋ฅผ ์ž๋™ํ™”ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

This model automatically detects claims that require fact-checking in Korean sentences. It can identify verifiable claims in news articles, political debates, and social media posts, automating the first step of the fact-checking workflow.

  • Base Model: beomi/KcELECTRA-base-v2022
  • Task: Claim Detection (Checkworthy Sentence Classification)
  • Language: Korean (ํ•œ๊ตญ์–ด)
  • Labels:
    • 0: ํŒฉํŠธ์ฒดํฌ๊ฐ€ ๋ถˆํ•„์š”ํ•œ ๋ฌธ์žฅ (Non-checkworthy)
    • 1: ํŒฉํŠธ์ฒดํฌ๊ฐ€ ํ•„์š”ํ•œ ์ฃผ์žฅ (Checkworthy claim)

๋ชจ๋ธ ๋ชฉํ‘œ (Model Objective)

์ž…๋ ฅ๋œ ํ•œ๊ตญ์–ด ๋ฌธ์žฅ์„ ๋ถ„์„ํ•˜์—ฌ ๋‹ค์Œ์„ ํŒ๋‹จํ•ฉ๋‹ˆ๋‹ค:

  • ๊ฒ€์ฆ ๊ฐ€๋Šฅํ•œ ์‚ฌ์‹ค์  ์ฃผ์žฅ์ธ์ง€
  • ํŒฉํŠธ์ฒดํ‚น์ด ํ•„์š”ํ•œ ์ •๋„๋Š” ์–ผ๋งˆ๋‚˜ ๋˜๋Š”์ง€

This model analyzes Korean sentences to determine:

  • Whether they contain verifiable factual claims
  • The degree to which fact-checking is needed

ํŒฉํŠธ์ฒดํฌ๊ฐ€ ํ•„์š”ํ•œ ์ฃผ์žฅ์˜ ์˜ˆ์‹œ (Checkworthy Claim Examples)

โœ… Label 1 (Checkworthy):

  • "์ฒญ๋…„ ์‹ค์—…๋ฅ ์ด ์ง€๋‚œ 3๋…„๊ฐ„ ๊ณ„์† ์ƒ์Šนํ–ˆ์Šต๋‹ˆ๋‹ค"
  • "์šฐ๋ฆฌ๋‚˜๋ผ GDP ์„ฑ์žฅ๋ฅ ์€ OECD ํ‰๊ท ์„ ๋„˜์–ด์„ฐ์Šต๋‹ˆ๋‹ค"
  • "์ด ์ •์ฑ…์œผ๋กœ ์ผ์ž๋ฆฌ๊ฐ€ 100๋งŒ ๊ฐœ ์ฐฝ์ถœ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค"

โŒ Label 0 (Non-checkworthy):

  • "์˜ค๋Š˜ ํ† ๋ก ํšŒ๋Š” SBS ์ƒ์•”๋™ ์ŠคํŠœ๋””์˜ค์—์„œ ์ง„ํ–‰ํ•˜๊ณ  ์žˆ๊ณ ์š”"
  • "๊ตญ๋ฏผ ์—ฌ๋Ÿฌ๋ถ„๊ป˜ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค"
  • "์ œ ์ƒ๊ฐ์—๋Š” ์ด ์ •์ฑ…์ด ์ข‹์€ ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค"

๋ฐ์ดํ„ฐ์…‹ (Dataset)

๋ฐ์ดํ„ฐ ์ถœ์ฒ˜

  • Source: CLEF CheckThat! Lab 2024
  • Task: Task 1 - Check-Worthiness Estimation
  • Original Dataset: English political debates and speeches
  • Translation: Machine-translated to Korean for training

๋ฐ์ดํ„ฐ์…‹ ํฌ๊ธฐ

  • Training Set: 22,501 samples
  • Validation Set: 1,032 samples
  • Test Set: 318 samples

๋ฐ์ดํ„ฐ ํŠน์„ฑ

  • ์ •์น˜ ํ† ๋ก , ์—ฐ์„ค๋ฌธ, ๋‰ด์Šค ๊ธฐ์‚ฌ์—์„œ ์ถ”์ถœ๋œ ๋ฌธ์žฅ
  • ์ „๋ฌธ ํŒฉํŠธ์ฒด์ปค๋“ค์ด ๋ ˆ์ด๋ธ”๋งํ•œ ๊ณ ํ’ˆ์งˆ ๋ฐ์ดํ„ฐ
  • ํด๋ž˜์Šค ๋ถˆ๊ท ํ˜•: Label 0 (65%) vs Label 1 (35%)

ํ•™์Šต ์„ธ๋ถ€์‚ฌํ•ญ (Training Details)

ํ•™์Šต ํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ

  • Epochs: 5
  • Batch Size (Train): 32
  • Batch Size (Eval): 64
  • Learning Rate: 3e-05
  • Weight Decay: 0.01
  • Warmup Ratio: 0.1
  • Precision: BF16
  • Optimizer: adamw_torch_fused
  • Max Sequence Length: 128 tokens
  • Seed: 42

ํ•™์Šต ํ™˜๊ฒฝ

  • GPU: NVIDIA GeForce RTX 4090 (24GB)
  • Training Time: 1.87 minutes
  • Framework: Hugging Face Transformers
  • Early Stopping: Patience 3 (based on F1 score)

์„ฑ๋Šฅ (Performance)

Validation Metrics

  • Accuracy: 97.58%
  • F1 Score: 94.80%
  • Precision: 93.83%
  • Recall: 95.80%

Test Metrics

  • Accuracy: 89.31%
  • F1 Score: 82.65%
  • Precision: 92.05%
  • Recall: 75.00%

Confusion Matrix (Test Set)

           Predicted
           0      1
Actual 0   203    7    (96.7% ์ •ํ™•๋„)
       1    27    81   (75.0% ์žฌํ˜„์œจ)

์„ฑ๋Šฅ ํ•ด์„:

  • ๋†’์€ Precision (92.05%): ๋ชจ๋ธ์ด "checkworthy"๋ผ๊ณ  ์˜ˆ์ธกํ•œ ๋ฌธ์žฅ์˜ 92%๊ฐ€ ์‹ค์ œ๋กœ ํŒฉํŠธ์ฒดํฌ๊ฐ€ ํ•„์š”
  • ์ ์ ˆํ•œ Recall (75.00%): ์‹ค์ œ checkworthy ๋ฌธ์žฅ์˜ 75%๋ฅผ ํƒ์ง€
  • ๋‚ฎ์€ False Positive (7๊ฐœ): ๋ถˆํ•„์š”ํ•œ ํŒฉํŠธ์ฒดํฌ ์š”์ฒญ ์ตœ์†Œํ™”

์‚ฌ์šฉ ๋ฐฉ๋ฒ• (How to Use)

1. ์„ค์น˜ (Installation)

pip install transformers torch

2. ๋ชจ๋ธ ๋กœ๋“œ (Loading the Model)

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# ๋ชจ๋ธ ๋กœ๋“œ
model_name = "jonghhhh/claim_factcheck"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# GPU ์‚ฌ์šฉ (์„ ํƒ์‚ฌํ•ญ)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval()

print(f"โœ… ๋ชจ๋ธ ๋กœ๋“œ ์™„๋ฃŒ! (Device: {device})")

3. ์ถ”๋ก  ์˜ˆ์‹œ (Inference Example)

๋‹จ์ผ ๋ฌธ์žฅ ๋ถ„๋ฅ˜

def predict_claim(text):
    """
    ์ž…๋ ฅ ๋ฌธ์žฅ์ด ํŒฉํŠธ์ฒดํฌ๊ฐ€ ํ•„์š”ํ•œ ์ฃผ์žฅ์ธ์ง€ ํŒ๋‹จํ•ฉ๋‹ˆ๋‹ค.

    Args:
        text (str): ๋ถ„์„ํ•  ํ•œ๊ตญ์–ด ๋ฌธ์žฅ

    Returns:
        dict: {
            'text': ์ž…๋ ฅ ๋ฌธ์žฅ,
            'is_checkworthy': True/False,
            'confidence': 0.0~1.0 (ํ™•์‹ ๋„),
            'label': 0 ๋˜๋Š” 1,
            'probabilities': {'non_checkworthy': 0.xx, 'checkworthy': 0.xx}
        }
    """
    # ํ† ํฌ๋‚˜์ด์ง•
    inputs = tokenizer(
        text,
        truncation=True,
        max_length=128,
        return_tensors="pt"
    )
    inputs = {k: v.to(device) for k, v in inputs.items()}

    # ์ถ”๋ก 
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=-1)
        predicted_label = torch.argmax(probs, dim=-1).item()
        confidence = probs[0][predicted_label].item()

    return {
        'text': text,
        'is_checkworthy': bool(predicted_label),
        'confidence': confidence,
        'label': predicted_label,
        'probabilities': {
            'non_checkworthy': probs[0][0].item(),
            'checkworthy': probs[0][1].item()
        }
    }

# ์‚ฌ์šฉ ์˜ˆ์‹œ
examples = [
    "์˜ค๋Š˜ ํ† ๋ก ํšŒ๋Š” SBS ์ƒ์•”๋™ ์ŠคํŠœ๋””์˜ค์—์„œ ์ง„ํ–‰ํ•˜๊ณ  ์žˆ๊ณ ์š”.",
    "์ฒญ๋…„ ์‹ค์—…๋ฅ ์ด ์ตœ๊ทผ 3๋…„๊ฐ„ ๊ณ„์† ์ƒ์Šนํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.",
    "์šฐ๋ฆฌ๋‚˜๋ผ GDP ์„ฑ์žฅ๋ฅ ์€ OECD ํ‰๊ท ์„ ๋„˜์–ด์„ฐ์Šต๋‹ˆ๋‹ค.",
    "๊ตญ๋ฏผ ์—ฌ๋Ÿฌ๋ถ„๊ป˜ ์ง„์‹ฌ์œผ๋กœ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค."
]

for text in examples:
    result = predict_claim(text)
    print(f"\n๐Ÿ“ ์ž…๋ ฅ: {result['text']}")
    print(f"{'๐Ÿ” ํŒฉํŠธ์ฒดํฌ ํ•„์š”' if result['is_checkworthy'] else 'โœ… ํŒฉํŠธ์ฒดํฌ ๋ถˆํ•„์š”'}")
    print(f"ํ™•์‹ ๋„: {result['confidence']*100:.1f}%")
    print(f"์ƒ์„ธ ํ™•๋ฅ : Non-CW {result['probabilities']['non_checkworthy']*100:.1f}% | CW {result['probabilities']['checkworthy']*100:.1f}%")

์ถœ๋ ฅ ์˜ˆ์‹œ:

๐Ÿ“ ์ž…๋ ฅ: ์ฒญ๋…„ ์‹ค์—…๋ฅ ์ด ์ตœ๊ทผ 3๋…„๊ฐ„ ๊ณ„์† ์ƒ์Šนํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
๐Ÿ” ํŒฉํŠธ์ฒดํฌ ํ•„์š”
ํ™•์‹ ๋„: 94.3%
์ƒ์„ธ ํ™•๋ฅ : Non-CW 5.7% | CW 94.3%

๐Ÿ“ ์ž…๋ ฅ: ์˜ค๋Š˜ ํ† ๋ก ํšŒ๋Š” SBS ์ƒ์•”๋™ ์ŠคํŠœ๋””์˜ค์—์„œ ์ง„ํ–‰ํ•˜๊ณ  ์žˆ๊ณ ์š”.
โœ… ํŒฉํŠธ์ฒดํฌ ๋ถˆํ•„์š”
ํ™•์‹ ๋„: 98.2%
์ƒ์„ธ ํ™•๋ฅ : Non-CW 98.2% | CW 1.8%

๋ฐฐ์น˜ ์ฒ˜๋ฆฌ (Batch Processing)

def predict_claims_batch(texts, batch_size=32):
    """
    ์—ฌ๋Ÿฌ ๋ฌธ์žฅ์„ ๋ฐฐ์น˜๋กœ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค.

    Args:
        texts (list): ๋ฌธ์žฅ ๋ฆฌ์ŠคํŠธ
        batch_size (int): ๋ฐฐ์น˜ ํฌ๊ธฐ

    Returns:
        list: ๊ฐ ๋ฌธ์žฅ์˜ ์˜ˆ์ธก ๊ฒฐ๊ณผ ๋ฆฌ์ŠคํŠธ
    """
    results = []

    for i in range(0, len(texts), batch_size):
        batch_texts = texts[i:i+batch_size]

        # ๋ฐฐ์น˜ ํ† ํฌ๋‚˜์ด์ง•
        inputs = tokenizer(
            batch_texts,
            truncation=True,
            max_length=128,
            padding=True,
            return_tensors="pt"
        )
        inputs = {k: v.to(device) for k, v in inputs.items()}

        # ๋ฐฐ์น˜ ์ถ”๋ก 
        with torch.no_grad():
            outputs = model(**inputs)
            probs = torch.softmax(outputs.logits, dim=-1)
            predicted_labels = torch.argmax(probs, dim=-1).cpu().numpy()

        # ๊ฒฐ๊ณผ ์ €์žฅ
        for j, text in enumerate(batch_texts):
            results.append({
                'text': text,
                'is_checkworthy': bool(predicted_labels[j]),
                'confidence': probs[j][predicted_labels[j]].item(),
                'label': int(predicted_labels[j])
            })

    return results

# ๋ฐฐ์น˜ ์ถ”๋ก  ์˜ˆ์‹œ
texts = [
    "๊ตญํšŒ์˜์› ์ •์›์„ 300๋ช…์œผ๋กœ ํ™•๋Œ€ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.",
    "๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.",
    "2024๋…„ ๊ฒฝ์ œ์„ฑ์žฅ๋ฅ ์ด 2.1%๋ฅผ ๊ธฐ๋กํ–ˆ์Šต๋‹ˆ๋‹ค.",
    # ... ๋” ๋งŽ์€ ๋ฌธ์žฅ๋“ค
]

batch_results = predict_claims_batch(texts)
checkworthy_claims = [r for r in batch_results if r['is_checkworthy']]
print(f"โœ… ์ด {len(texts)}๊ฐœ ๋ฌธ์žฅ ์ค‘ {len(checkworthy_claims)}๊ฐœ๊ฐ€ ํŒฉํŠธ์ฒดํฌ ํ•„์š”")

4. ์‹ค์ „ ํ™œ์šฉ ์˜ˆ์‹œ (Real-world Use Case)

# ๋‰ด์Šค ๊ธฐ์‚ฌ์—์„œ ํŒฉํŠธ์ฒดํฌ ๋Œ€์ƒ ์ถ”์ถœ
def extract_checkworthy_claims(article_text, threshold=0.7):
    """
    ๊ธฐ์‚ฌ์—์„œ ํŒฉํŠธ์ฒดํฌ๊ฐ€ ํ•„์š”ํ•œ ๋ฌธ์žฅ๋“ค์„ ์ถ”์ถœํ•ฉ๋‹ˆ๋‹ค.

    Args:
        article_text (str): ๋‰ด์Šค ๊ธฐ์‚ฌ ์ „๋ฌธ
        threshold (float): checkworthy ํŒ๋‹จ ์ž„๊ณ„๊ฐ’ (0.0~1.0)

    Returns:
        list: ํŒฉํŠธ์ฒดํฌ ๋Œ€์ƒ ๋ฌธ์žฅ๋“ค
    """
    # ๋ฌธ์žฅ ๋ถ„๋ฆฌ (๊ฐ„๋‹จํ•œ ์˜ˆ์‹œ)
    sentences = [s.strip() for s in article_text.split('.') if s.strip()]

    # ๋ฐฐ์น˜ ์˜ˆ์ธก
    results = predict_claims_batch(sentences)

    # ์ž„๊ณ„๊ฐ’ ์ด์ƒ์˜ checkworthy ๋ฌธ์žฅ๋งŒ ํ•„ํ„ฐ๋ง
    checkworthy_claims = [
        r for r in results
        if r['is_checkworthy'] and r['confidence'] >= threshold
    ]

    # ํ™•์‹ ๋„ ์ˆœ์œผ๋กœ ์ •๋ ฌ
    checkworthy_claims.sort(key=lambda x: x['confidence'], reverse=True)

    return checkworthy_claims

# ์‚ฌ์šฉ ์˜ˆ์‹œ
article = """
์ •๋ถ€๋Š” ์˜ค๋Š˜ ๊ฒฝ์ œ์ •์ฑ… ๋ฐฉํ–ฅ์„ ๋ฐœํ‘œํ–ˆ์Šต๋‹ˆ๋‹ค.
์ฒญ๋…„ ์‹ค์—…๋ฅ ์ด ์ง€๋‚œํ•ด ๋Œ€๋น„ 2.3%p ๊ฐ์†Œํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”์Šต๋‹ˆ๋‹ค.
์ด๋Š” ์—ญ๋Œ€ ์ตœ๋Œ€ ํญ์˜ ํ•˜๋ฝ์ž…๋‹ˆ๋‹ค.
์•ž์œผ๋กœ๋„ ์ผ์ž๋ฆฌ ์ฐฝ์ถœ์— ํž˜์“ฐ๊ฒ ๋‹ค๊ณ  ๊ฐ•์กฐํ–ˆ์Šต๋‹ˆ๋‹ค.
"""

claims = extract_checkworthy_claims(article, threshold=0.8)
print(f"๐Ÿ” ๋ฐœ๊ฒฌ๋œ ํŒฉํŠธ์ฒดํฌ ๋Œ€์ƒ: {len(claims)}๊ฐœ\n")

for i, claim in enumerate(claims, 1):
    print(f"{i}. {claim['text']}")
    print(f"   ํ™•์‹ ๋„: {claim['confidence']*100:.1f}%\n")

๋ชจ๋ธ ์•„ํ‚คํ…์ฒ˜ (Model Architecture)

  • Model Type: ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately)
  • Hidden Size: 768
  • Number of Layers: 12
  • Number of Attention Heads: 12
  • Vocabulary Size: 32,000
  • Max Sequence Length: 128 tokens
  • Classification Head: Linear layer (768 โ†’ 2)

ํ•œ๊ณ„ ๋ฐ ๊ณ ๋ ค์‚ฌํ•ญ (Limitations)

  1. ๋„๋ฉ”์ธ ํŠนํ™”: ์ •์น˜/๋‰ด์Šค ๋„๋ฉ”์ธ์— ์ตœ์ ํ™”๋˜์–ด ์žˆ์–ด, ์ผ์ƒ ๋Œ€ํ™”๋‚˜ ๊ธฐ์ˆ  ๋ฌธ์„œ์—๋Š” ์„ฑ๋Šฅ์ด ๋–จ์–ด์งˆ ์ˆ˜ ์žˆ์Œ
  2. ๊ธธ์ด ์ œํ•œ: ์ตœ๋Œ€ 128 ํ† ํฐ๊นŒ์ง€๋งŒ ์ฒ˜๋ฆฌ ๊ฐ€๋Šฅ (์•ฝ 100-150 ๋‹จ์–ด)
  3. ๊ธฐ๊ณ„ ๋ฒˆ์—ญ ๋ฐ์ดํ„ฐ: ์˜์–ด์—์„œ ๋ฒˆ์—ญ๋œ ๋ฐ์ดํ„ฐ๋กœ ํ•™์Šต๋˜์–ด ์ž์—ฐ์Šค๋Ÿฌ์šด ํ•œ๊ตญ์–ด ํ‘œํ˜„์—์„œ ์„ฑ๋Šฅ ์ฐจ์ด ๊ฐ€๋Šฅ
  4. ์ด์ง„ ๋ถ„๋ฅ˜: Checkworthy ์ •๋„๋ฅผ 0/1๋กœ๋งŒ ๋ถ„๋ฅ˜ (์„ธ๋ฐ€ํ•œ ์ ์ˆ˜ ์ œ๊ณต ์•ˆ ํ•จ)
  5. False Negative: ์‹ค์ œ ์ฃผ์žฅ์˜ 25%๋ฅผ ๋†“์น  ์ˆ˜ ์žˆ์Œ (Recall 75%)

๊ฐœ์„  ๋ฐฉํ–ฅ (Future Improvements)

  • ํ•œ๊ตญ์–ด ๋„ค์ดํ‹ฐ๋ธŒ ํŒฉํŠธ์ฒดํฌ ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ์ถ”๊ฐ€ ํ•™์Šต
  • ๋” ๊ธด ๋ฌธ๋งฅ ์ฒ˜๋ฆฌ๋ฅผ ์œ„ํ•œ ๋ชจ๋ธ ์—…๊ทธ๋ ˆ์ด๋“œ (max_length 256+)
  • ๋‹ค์ค‘ ํด๋ž˜์Šค ๋ถ„๋ฅ˜ (checkworthy ์ ์ˆ˜๋ฅผ 0-5 ์ฒ™๋„๋กœ)
  • ์ฃผ์žฅ์˜ ์ฃผ์ œ ์นดํ…Œ๊ณ ๋ฆฌ ๋ถ„๋ฅ˜ ๊ธฐ๋Šฅ ์ถ”๊ฐ€

๋ผ์ด์„ ์Šค (License)

์ด ๋ชจ๋ธ์€ ๋ฒ ์ด์Šค ๋ชจ๋ธ์ธ beomi/KcELECTRA-base-v2022์˜ ๋ผ์ด์„ ์Šค๋ฅผ ๋”ฐ๋ฆ…๋‹ˆ๋‹ค.

์ธ์šฉ (Citation)

์ด ๋ชจ๋ธ์„ ์—ฐ๊ตฌ๋‚˜ ํ”„๋กœ์ ํŠธ์— ์‚ฌ์šฉํ•˜์‹ ๋‹ค๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ธ์šฉํ•ด์ฃผ์„ธ์š”:

@misc{korean-claim-factcheck-2025,
  author = {Jonghhhh},
  title = {Korean Claim Detection Model for Fact-Checking},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/jonghhhh/claim_factcheck}},
  note = {Based on KcELECTRA-base-v2022}
}

์ฐธ๊ณ  ์ž๋ฃŒ (References)

์—ฐ๋ฝ์ฒ˜ (Contact)

์งˆ๋ฌธ์ด๋‚˜ ํ”ผ๋“œ๋ฐฑ์ด ์žˆ์œผ์‹œ๋ฉด Issues๋ฅผ ํ†ตํ•ด ๋‚จ๊ฒจ์ฃผ์„ธ์š”!


Tags: claim-detection, fact-checking, korean, electra, text-classification, checkworthy, misinformation-detection

Downloads last month
7
Safetensors
Model size
0.1B params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support