Triple Omission
A category of factual error in knowledge-graph-to-text tasks in which a triple from the input graph is not mentioned at all in the generated text, constituting an information-missing type of error.
Timeline (last 90 days)
On graph-grounded metrics, LLMs perform better than fine-tuned models in reducing contradictions and triple omissions.
Triple omission means that a triple in the input graph is not mentioned at all in the generated text; contradiction means that the generated text expresses content that conflicts with the input graph.
All Facts (2)
On graph-grounded metrics, LLMs perform better than fine-tuned models in reducing contradictions and triple omissions.
50%UnverifiedTriple omission means that a triple in the input graph is not mentioned at all in the generated text; contradiction means that the generated text expresses content that conflicts with the input graph.
50%