SummaC

Objective:

SummaC evaluates summarization models by assessing factual consistency. It uses Natural Language Inference (NLI) techniques to determine if the generated summary logically entails, contradicts, or is neutral with respect to the source text. This approach allows for the detection of factual hallucinations or omissions in the generated summary, focusing on maintaining logical consistency with the original input. SummaC is commonly used for fact-checking in summarization tasks to ensure accurate information representation.

Required Columns in Dataset:

LLM Summary, Original Document

Interpretation:

• High SummaC: Indicates that the summary is factually consistent with the source, with little to no contradictions or hallucinations.

• Low SummaC: Reflects potential factual inconsistencies or logical contradictions between the summary and the original text.

Execution via UI:

Execution via SDK:

metrics=[
    {"name": "SummaC", "config": {"model": "gpt-4o-mini", "provider": "openai"}, "column_name": "your-text", "schema_mapping": schema_mapping}
]