FActScore
An evaluation method for assessing the factual accuracy of content generated by language models. It decomposes generated text into atomic facts and verifies each one individually against a knowledge source, producing a scored assessment of factual correctness at a granular level.
Timeline (last 90 days)
CSR (Citation Survival Rate) requires each citation to pass two checks: a support check that uses FActScore-style atomic claim decomposition plus an ALCE-style entailment model to verify whether the cited block supports the claim, and a provenance check that relies on tool-call logs to confirm the URL actually appeared in the content the Agent retrieved.
The concept of atomic facts originated in the fields of knowledge graphs and information extraction, and has been widely adopted in recent years in RAG systems and LLM evaluation (such as the FActScore method).
AI output quality inspection mechanisms that have emerged include tools such as FActScore, GPTZero, and Originality.ai.
All Facts (3)
CSR (Citation Survival Rate) requires each citation to pass two checks: a support check that uses FActScore-style atomic claim decomposition plus an ALCE-style entailment model to verify whether the cited block supports the claim, and a provenance check that relies on tool-call logs to confirm the URL actually appeared in the content the Agent retrieved.
50%UnverifiedThe concept of atomic facts originated in the fields of knowledge graphs and information extraction, and has been widely adopted in recent years in RAG systems and LLM evaluation (such as the FActScore method).
50%UnverifiedAI output quality inspection mechanisms that have emerged include tools such as FActScore, GPTZero, and Originality.ai.
50%