CiteVQA: A Benchmark for Document Visual Question Answering with Evidence Attribution

CiteVQA is a document visual question answering benchmark designed to evaluate faithful evidence attribution. The dataset contains 1,897 question-answer pairs grounded in real-world PDF documents. It was created by opendatalab and last updated on 2026-05-13.

Use Cases

• Train models for document visual question answering based on the benchmark's question-answer pairs.
• Evaluate a system's ability to cite supporting evidence based on the requirement for element-level grounding.
• Benchmark multimodal AI performance on long, real-world PDFs as described in the dataset's purpose.

Strengths

• Contains 1,897 question-answer pairs for evaluation.
• Specifically designed for element-level evidence grounding in documents.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

opendatalab

Freshness

Last updated 2026-05-13 08:58:02; freshness should be verified.