Biomed-Enriched: PubMed Articles Annotated by LLM for Biomedical NLP

A biomedical dataset derived from PubMed and enriched through a two-stage annotation process. Llama 3.1 70B Instruct annotated 400,000 paragraphs for document type, domain, and educational quality. The dataset was created by researchers from Sorbonne Université and INRIA Paris, with a last update recorded on 2025-06-27.

Use Cases

• Pretraining language models on biomedical text based on the PubMed-derived content.
• Extracting rare or hidden biomedical concepts based on the LLM-generated annotations.
• Classifying biomedical literature by document type based on the annotated labels.
• Assessing the educational quality of biomedical text passages based on the provided annotations.
• Training models for domain-specific biomedical NLP tasks based on the enriched metadata.

Strengths

• Contains 400,000 annotated paragraphs, providing a substantial base for training.
• Annotations include document type, domain, and educational quality, offering multi-faceted metadata.
• Uses a large language model (Llama 3.1 70B Instruct) for annotation, suggesting a degree of automation and scale.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown beyond the 400K annotated paragraphs, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

PubMed

Collection Method

A two-stage annotation process where an LLM annotated paragraphs, with subsequent processing.

Freshness

Last updated 2025-06-27 10:29:46; freshness should be verified.