In the KGHaluBench entity-level filter, a threshold of 0.750 achieved the highest alignment with human judges at 79.19%, but resulted in a lower recall of 73.17%, indicating the filter was overly strict.

Authors

• Person: Not available Organization: arXiv
  A Knowledge Graph-Based Hallucination Benchmark for Evaluating ...

Sources

• A Knowledge Graph-Based Hallucination Benchmark for Evaluating ... arxiv.org arXiv via serper

Referenced by nodes (2)

• KGHaluBench concept
• entity-level filtering concept