llm-semantic-router/fact-check-classification-dataset

Fact-Check Classification Dataset

🎯 Purpose: Binary classification dataset for determining whether a prompt needs external fact-checking.

Dataset Description

This dataset is designed to train classifiers that can route LLM requests based on whether they require external fact verification. It's part of the vLLM Semantic Router project.

Labels

• FACT_CHECK_NEEDED (1): Information-seeking questions requiring external verification

  • Factual questions about dates, people, places, events
  • Questions about current events or statistics
  • Claims that can be verified against external sources

• NO_FACT_CHECK_NEEDED (0): Prompts that don't need external fact verification

  • Creative writing requests
  • Coding/programming questions
  • Mathematical calculations
  • Opinion or advice requests
  • Summarization tasks

Dataset Sources

FACT_CHECK_NEEDED Sources

Source	Description	Reference
NISQ (ISQ)	Gold standard Information-Seeking Questions	LREC 2024
FactCHD	Fact-conflicting hallucination detection queries	Chen et al., 2024
FaithDial	Information-seeking dialogue questions	TACL 2022
HaluEval	QA questions from hallucination benchmark	ACL EMNLP 2023
SQuAD	Reading comprehension questions	Stanford
TruthfulQA	Questions about common misconceptions	Lin et al.
TriviaQA	Trivia questions	Joshi et al.
HotpotQA	Multi-hop factual questions	Yang et al.
RAG Dataset	Questions for retrieval-augmented generation	neural-bridge

NO_FACT_CHECK_NEEDED Sources

Source	Description	Reference
NISQ (Non-ISQ)	Non-Information-Seeking Questions (deliberative, rhetorical)	LREC 2024
WritingPrompts	Creative writing prompts from Reddit	euclaise
Dolly	Creative writing, brainstorming, opinions	Databricks
Alpaca	Non-factual instructions (coding, creative, math)	Stanford

Dataset Structure

{
    "text": "When was the Eiffel Tower built?",
    "label": "FACT_CHECK_NEEDED",
    "label_id": 1,
    "source": "TriviaQA",
    "original_label": "trivia_question"
}

Fields

• text (string): The input prompt/question
• label (string): Binary classification label
• label_id (int): Numeric label (0 or 1)
• source (string): Original dataset source
• original_label (string): Original label from source dataset

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("llm-semantic-router/fact-check-classification-dataset")

# Access splits
train_data = dataset["train"]
val_data = dataset["validation"]
test_data = dataset["test"]

# Example usage
for example in train_data.select(range(5)):
    print(f"Text: {example['text'][:50]}...")
    print(f"Label: {example['label']}")
    print()

Statistics

Split	Examples	FACT_CHECK_NEEDED	NO_FACT_CHECK_NEEDED
Train	15,624	~10,492	~5,132
Validation	3,348	~2,248	~1,100
Test	3,349	~2,248	~1,101

Related Models

• llm-semantic-router/mmbert-fact-check-merged - Multilingual fact-check classifier

Citation

If you use this dataset, please cite the relevant source datasets:

@inproceedings{sun2024nisq,
  title={What Are the Implications of Your Question? Non-Information Seeking Question-Type Identification},
  author={Sun, Yao and others},
  booktitle={LREC-COLING},
  year={2024}
}

@article{dziri2022faithdial,
  title={FaithDial: A Faithful Benchmark for Information-Seeking Dialogue},
  author={Dziri, Nouha and others},
  journal={TACL},
  year={2022}
}

@article{chen2024factchd,
  title={FactCHD: Benchmarking Fact-Conflicting Hallucination Detection},
  author={Chen, Xiang and others},
  year={2024}
}

License

Apache 2.0 - See individual source datasets for their specific licenses.

Contact

• Organization: llm-semantic-router
• Project: vLLM Semantic Router