WikiProfile: 21,500 Factual Knowledge Questions for Language Model Evaluation

21,500 question instances derived from 2,150 facts extracted from English Wikipedia summaries. The dataset was created by Google and last updated on Hugging Face in June 2026. Each fact is structured as a proposition between a subject and an object entity.

Use Cases

• Benchmarking language model performance on factual recall based on structured Wikipedia propositions.
• Training or fine-tuning models for knowledge-intensive question answering tasks.
• Analyzing model biases or gaps in factual knowledge grounded in Wikipedia summaries.

Strengths

• Contains 21,500 distinct question instances, providing a substantial testbed.
• Based on 2,150 facts, each grounded in the first paragraph of a Wikipedia page.
• Structured with explicit subject and object entities for precise evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect the temporal and editorial bias inherent to Wikipedia as a source.

Provenance

Source

Google

Collection Method

Facts extracted from the first paragraph (summary) of English Wikipedia pages.

Freshness

Last updated 2026-06-19 09:58:02; freshness should be verified.