ArtemkaT08/medical-shorts-silver-claims-benchmark

Medical Shorts Silver Claims Benchmark

This dataset contains silver labels for evaluating medical claim detection and verification in short-form YouTube videos.

The dataset does not include video files, audio files, thumbnails, or keyframes. It only includes YouTube video IDs, metadata, normalized medical claims, silver labels, and evidence source references.

Dataset Size

• —Videos: 236
• —Silver claims: 470
• —Visual-dependent claims: 22
• —Videos with medical claims: 205
• —Videos without medical claims: 31
• —Risk levels: low=138, medium=50, high=48

Intended Use

This dataset is intended for research and benchmarking of LLM-based medical claim detection and verification systems.

Load

from datasets import load_dataset

dataset = load_dataset(
    "ArtemkaT08/medical-shorts-silver-claims-benchmark",
    token=True,
)

Labels

Labels were generated by a stronger LLM and should be treated as silver labels, not expert medical annotations.

License

The dataset card currently uses license: other until a final release license is selected.

Limitations

• —Labels were not created by medical professionals.
• —Some source retrieval may be incomplete.
• —Some videos may become unavailable.
• —Transcription and keyframe extraction errors may affect annotation quality.
• —This dataset should not be used for clinical decision-making.

Schema

Each row represents one YouTube video and includes:

• —video_id
• —duration_sec
• —language
• —contains_medical_claim
• —has_visual_dependent_claim
• —up to 3 claim annotations
• —modality labels
• —consensus category
• —evidence coverage score
• —plain-English explanation
• —source references
• —confidence and quality flags