Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED).

Year:	2025
Title:	Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED).
Abstract:	Introduces MAKGED, a multi-agent LLM+GCN approach for KG error detection. Four agents (head-forward/backward, tail-forward/backward) are trained on bidirectional subgraph embeddings concatenated with LLM query embeddings. They discuss and vote on each triple’s correctness. Outperforms SOTA by +0.73% on FB15k and +6.62% on WN18RR.
Objectives:	Enhance triple error detection by combining structured graph views with LLM reasoning and multi-agent consensus.
Methodology:	Build bidirectional subgraphs for head/tail of each triple; use GCN to embed structure and Llama2 for semantic embedding; train four specialized agents; agents independently evaluate triples then engage in multi-round discussion and vote on final decision.
Algorithm Used:	MAKGED (LLM+GCN multi-agent).
Top Model:	Multi-agent (LLM + GCN).
Accuracy:	Achieves 0.73% (FB15k) and 6.62% (WN18RR) higher accuracy than previous best methods.
Advantages:	Integrates fine-grained graph info with LLM knowledge; multi-agent discussion yields transparent, robust decisions; strong gains on benchmarks.
Drawbacks:	High computational and implementation complexity; requires multiple LLMs/agents; performance depends on quality of subgraphs and LLM reasoning; novel (practical impact TBD).
Source:	https://www.researchgate.net/publication/388422696_Harnessing_Diverse_Perspectives_A_Multi-Agent_Framework_for_Enhanced_Error_Detection_in_Knowledge_Graphs Edit

Criteria (6)

Name	Description	Definitions
Accuracy	In the framework Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED), the criterion "Accuracy" …	Quality dimension from Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge … Introduces MAKGED, a multi-agent LLM+GCN approach for KG error detection. Four agents (head-forward/backward, tail-forward/backward) are …
Objectivity (Consensus	In the framework Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED), the criterion of …	Introduces MAKGED, a multi-agent LLM+GCN approach for KG error detection. Four agents (head-forward/backward, tail-forward/backward) are …
Objectivity (Consensus)	Quality dimension from Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED).	Quality dimension from Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge …
Interpretability	In the framework Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED), the criterion of …	Quality dimension from Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge … Introduces MAKGED, a multi-agent LLM+GCN approach for KG error detection. Four agents (head-forward/backward, tail-forward/backward) are …
Reliability	Evaluates how trustworthy and dependable the knowledge graph is as a source of information.	Quality dimension from Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge … Introduces MAKGED, a multi-agent LLM+GCN approach for KG error detection. Four agents (head-forward/backward, tail-forward/backward) are …
Consistent representation	In the framework Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge Graphs (MAKGED), the criterion "Consistent …	Quality dimension from Harnessing Diverse Perspectives: A Multi-Agent Framework for Enhanced Error Detection in Knowledge … Introduces MAKGED, a multi-agent LLM+GCN approach for KG error detection. Four agents (head-forward/backward, tail-forward/backward) are …