Knowledge Tree
GraphEval
Facts (18)
Sources
A Knowledge-Graph Based LLM Hallucination Evaluation Framework themoonlight.io 8 facts
procedureThe 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.
claimGraphEval improves balanced accuracy in hallucination detection when used with various Natural Language Inference (NLI) models.
claimGraphEval utilizes a structured knowledge graph approach to provide higher hallucination detection accuracy and to explain the specific locations of inaccuracies within Large Language Model outputs.
claimThe research paper evaluates the effectiveness of the GraphEval and GraphCorrect frameworks using three established benchmarks: SummEval, QAGS-C, and QAGS-X.
claimThe integration of GraphCorrect with GraphEval provides a methodology for rectifying hallucinations in Large Language Model outputs, with potential applications in fields requiring factual correctness such as medical advice or legal documentation.
procedureThe GraphEval framework constructs a Knowledge Graph from LLM output through a four-step pipeline: (1) processing input text, (2) detecting unique entities, (3) performing coreference resolution to retain only specific references, and (4) extracting relations to form triples of (entity1, relation, entity2).
claimThe authors of the GraphEval framework focus on detecting hallucinations within a defined context rather than identifying discrepancies between LLM responses and broader training data.
claimThe GraphEval framework categorizes an entire LLM output as containing a hallucination if at least one triple within the constructed Knowledge Graph is flagged as inconsistent with the provided context.
EdinburghNLP/awesome-hallucination-detection - GitHub github.com 2 facts
procedureThe GraphEval and GraphCorrect framework detects hallucinations by extracting knowledge graph triples from LLM output and comparing their entailment against provided context, then corrects them by prompting an LLM to generate factually correct triples and replacing the non-factual information.
claimThe GraphEval and GraphCorrect framework utilizes HHEM (DeBERTaV3), TRUE, and TrueTeacher (T5-XXL) as underlying NLI models and Claude2 as the underlying LLM for experiments.
A Knowledge-Graph Based LLM Hallucination Evaluation Framework arxiv.org Jul 15, 2024 2 facts
claimGraphEval identifies specific triples within a Knowledge Graph that are prone to hallucinations, providing insight into the location of hallucinations within an LLM response.
measurementUsing GraphEval in conjunction with state-of-the-art natural language inference (NLI) models improves balanced accuracy on various hallucination benchmarks compared to using raw NLI models alone.
A knowledge-graph based LLM hallucination evaluation framework amazon.science 2 facts
A Knowledge Graph-Based Hallucination Benchmark for Evaluating ... arxiv.org Feb 23, 2026 1 fact
referenceGraphEval (Liu et al., 2024) and Dynamic-KGQA (Dammu et al., 2025) dynamically construct test datasets to remain up-to-date.
A Knowledge-Graph Based LLM Hallucination Evaluation Framework semanticscholar.org 1 fact
claimGraphEval is a hallucination evaluation framework for Large Language Models that represents information using Knowledge Graph structures, as presented in the paper 'A Knowledge-Graph Based LLM Hallucination Evaluation Framework' by Sansford and Richardson.
Practices, opportunities and challenges in the fusion of knowledge ... frontiersin.org 1 fact
referenceKG-FIT (Jiang P. et al., 2024) and GraphEval (Sansford et al., 2024) are modular frameworks that inject knowledge graph-derived signals during fine-tuning or evaluation to make large language models more robust, verifiable, and explainable in knowledge-intensive tasks.
Unknown source 1 fact
claimGraphEval is a knowledge-graph based LLM hallucination evaluation framework.