The GraphEval framework detects hallucinations by using a pretrained Natural Language Inference (NLI) model to compare each triple in the constructed Knowledge Graph against the original context, flagging a triple as a hallucination if the NLI model predicts inconsistency with a probability score greater than 0.5.

Authors

• Person: Not available Organization: The Moonlight
  A Knowledge-Graph Based LLM Hallucination Evaluation Framework

Sources

• A Knowledge-Graph Based LLM Hallucination Evaluation Framework www.themoonlight.io The Moonlight via serper

Referenced by nodes (5)

• Knowledge Graph concept
• hallucination concept
• natural language inference (NLI) concept
• context concept
• GraphEval concept