NoAtmosphere0/LIAR_RAW_RAWFC_Merged

LIAR-RAW + RAWFC Merged Fact-Checking Dataset

This dataset is a unified resource for claim verification and fact-checking research, created by merging and harmonizing the LIAR-RAW and RAWFC datasets. It provides a large, diverse collection of real-world claims and supporting or refuting evidence, suitable for training and evaluating machine learning models in automated fact-checking and related tasks.

---

Dataset Contents

Each entry in the dataset contains:

• claim: The main statement or claim to be verified.
• label: The veracity label (e.g., true, false, half-true, etc.).
• explain: An explanation supporting the assigned label.
• reports: A list of evidence reports, each as a dictionary with:
  • report_id (if available)
  • content: The textual evidence.

Example:

{
  "claim": "The sky is green.",
  "label": "false",
  "explain": "Scientific consensus says the sky appears blue due to Rayleigh scattering.",
  "reports": [
    {"report_id": "1234567", "content": "A NASA article explains why the sky is blue."},
    {"report_id": "2345678", "content": "Physics textbook reference on atmospheric optics."}
  ]
}

---

Data Sources and Preprocessing

• LIAR-RAW:

  • Cleaned to standardize labels (mostly-true → true, barely-true/pants-fire → false).
  • Reports are simplified to retain only relevant fields.

• RAWFC:

  • Data is loaded from split folders and concatenated.
  • The label half is renamed to half-true.
  • Only the content of each evidence report is kept.

• Merged Dataset:

  • Train, validation, and test splits from both sources are merged.
  • Validation is combined with training for a larger training set.
  • Saved as train.json and test.json in the unified format.

---

Dataset Statistics

• Scale: Over 167,000 unique evidence reports in the training set.
• Splits: Provided as train.json and test.json (see Data Structure).
• Format: JSON, and optionally as HuggingFace DatasetDict.

---

Usage

The dataset is suitable for:

• Training and evaluating claim verification and fact-checking models.
• Research in evidence retrieval, claim matching, and misinformation detection.
• NLP tasks involving multi-document reasoning and textual entailment.

Loading Example (Python)

import json
with open('train.json', 'r') as f:
    data = json.load(f)
print(data[0])

---

Limitations

• Combined labels may lose some original nuance.
• Evidence reports vary in length, quality, and relevance.
• Dataset may still contain noisy or ambiguous examples, and label balance is not guaranteed.
• For detailed limitations of models trained on this data, see the main project README.

---

License & Citation

Please refer to the original licenses of LIAR-RAW and RAWFC.
If you use this merged dataset, cite the relevant original works and this repository.

---

Acknowledgments

• LIAR-RAW
• RAWFC: Raw Fact-Checking Dataset