KairosQA: Temporally Evolving Wikidata Facts for LLM Evaluation

KairosQA is a question-answering dataset designed to evaluate the temporal reasoning capabilities of Large Language Models. It focuses on subject–relation–object triplets from Wikidata that changed at least twice between 2018 and 2025. The dataset was created by author kyutai and was last updated on Hugging Face in May 2026.

Use Cases

• Benchmarking LLM performance on temporally evolving facts based on Wikidata triplets.
• Training models to track and reason about changes in structured knowledge over time.
• Evaluating the temporal alignment of model-generated answers against a ground truth timeline.
• Studying the impact of data temporality on model understanding and factuality.

Strengths

• Focuses on temporally evolving facts, a specific and challenging aspect of reasoning.
• Based on structured Wikidata triplets, providing a clear factual backbone.
• Covers a defined time range from 2018 to 2025 for temporal analysis.

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 temporal and source bias inherent to the underlying Wikidata extraction process.

Provenance

Source

Wikidata

Collection Method

Extracted subject–relation–object triplets that changed at least twice between 2018 and 2025.

Time Range

2018 to 2025

Freshness

Last updated 2026-05-26 14:04:57; freshness should be verified.