This paper introduces a novel benchmark dataset, ConceptEdit, for editing conceptual knowledge in large language models (LLMs). The study investigates the ability of LLMs to modify conceptual knowledge, which is a critical yet underexplored area. The research constructs ConceptEdit based on the DBpedia Ontology, a widely recognized and cross-domain ontology that preserves conceptual knowledge hierarchically. The dataset includes 452 concepts, 8,767 instances, and 22 superclasses. It also introduces two new metrics, Instance Change and Concept Consistency, to evaluate the effectiveness of existing editing methods on conceptual knowledge. The study evaluates four editing methods: FT, ROME, MEMIT, and PROMPT. The results show that while these methods can efficiently modify concept-level definitions, they may also distort related instance-level knowledge, leading to poor performance. The paper highlights the importance of developing more accurate metrics to assess the impact of conceptual knowledge editing on LLMs. It also discusses the challenges of editing conceptual knowledge, including the need for more precise evaluation metrics and the potential for conceptual knowledge to influence instance-level knowledge in LLMs. The study concludes that further research is needed to better understand how LLMs learn and update concepts, and to develop more effective methods for editing conceptual knowledge in LLMs.This paper introduces a novel benchmark dataset, ConceptEdit, for editing conceptual knowledge in large language models (LLMs). The study investigates the ability of LLMs to modify conceptual knowledge, which is a critical yet underexplored area. The research constructs ConceptEdit based on the DBpedia Ontology, a widely recognized and cross-domain ontology that preserves conceptual knowledge hierarchically. The dataset includes 452 concepts, 8,767 instances, and 22 superclasses. It also introduces two new metrics, Instance Change and Concept Consistency, to evaluate the effectiveness of existing editing methods on conceptual knowledge. The study evaluates four editing methods: FT, ROME, MEMIT, and PROMPT. The results show that while these methods can efficiently modify concept-level definitions, they may also distort related instance-level knowledge, leading to poor performance. The paper highlights the importance of developing more accurate metrics to assess the impact of conceptual knowledge editing on LLMs. It also discusses the challenges of editing conceptual knowledge, including the need for more precise evaluation metrics and the potential for conceptual knowledge to influence instance-level knowledge in LLMs. The study concludes that further research is needed to better understand how LLMs learn and update concepts, and to develop more effective methods for editing conceptual knowledge in LLMs.