Medical Question Answering Dataset With UMLS Annotations

A dataset for medical question answering, converted from MedQuAD. It contains English medical questions and answers with UMLS (Unified Medical Language System) annotations, including concept identifiers and semantic types. The dataset excludes 31,034 records from specific sources to respect copyright.

Use Cases

• Train a question-answering model on medical questions annotated with UMLS concept identifiers (CUI).
• Analyze the distribution of UMLS semantic types across different medical question categories.
• Build a system to map medical synonyms, using the 'synonyms' column data separated by '|' characters.

Strengths

• Includes UMLS annotations (CUI and semantic types) for structured medical knowledge.
• Excludes 31,034 records from copyrighted sources, clarifying permissible use.
• Last updated in December 2023.

Limitations

• Excludes a significant portion (31,034 records) of the original data, reducing scope.
• Column structure and total row count are unknown, hindering precise assessment.
• Potential for question type discrepancies as noted in the original description.

Provenance

Source

Converted from the MedQuAD dataset.

Collection Method

Conversion of an existing medical QA dataset; specific original gathering method unknown.

Freshness

Last updated on 2023-12-22.