SciClaimHunt: ~110,000 Scientific Claims with Evidence and Full Paper Text

SciClaimHunt is a large-scale scientific claim verification dataset comprising approximately 110,000 instances of scientific claims, supporting evidence, and full research paper text. It was created by AnshulS and last updated on the Hugging Face platform on May 6, 2025. The dataset enables experimentation on scientific fact verification and document-level reasoning.

Use Cases

• Train models for scientific claim verification based on the claim-evidence pairs.
• Benchmark evidence retrieval systems using the supporting evidence and full paper text.
• Develop models for document-level reasoning tasks based on the provided research paper context.

Strengths

• Large scale with approximately 110,000 instances.
• Includes three key components per instance: a claim, supporting evidence, and full research paper text.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

AnshulS on Hugging Face

Freshness

Last updated 2025-05-06 06:20:17; freshness should be verified.