notesum.ai

Published at October 23

GraphTeam: Facilitating Large Language Model-based Graph Analysis via Multi-Agent Collaboration

cs.CV

cs.AI

Released Date: October 23, 2024

Authors: Xin Li¹, Qizhi Chu¹, Yubin Chen², Yang Liu¹, Yaoqi Liu¹, Zekai Yu¹, Weize Chen³, Chen Qian³, Chuan Shi¹, Cheng Yang¹

Aff.: ¹Beijing University of Posts and Telecommunications; ²The Chinese University of Hong Kong; ³Tsinghua University

Arxiv: https://arxiv.org/abs/2410.18032v1

Refer to caption

Category	Models	Talk like a Graph	GraphWiz	NLGraph	LLM4DyG	GraphInstruct	GNN-AutoGL
Main	GPT-4o-mini	71.67	47.61	51.40	58.03	53.67	3.28
	SOTA	97.30	65.00	79.70	53.71	43.53	-
	GraphTeam	99.13	88.61	97.80	94.81	88.14	48.44
	- question agent	85.87	62.81	79.22	59.18	85.86	14.53
Input-Output Normalization	- answer agent	94.80	87.97	96.70	94.27	85.25	45.78
	- documentation	92.03	87.97	96.40	94.07	86.12	46.88
Knowledge Retrieval	- experience	89.96	82.83	54.70	60.26	76.64	4.76
	- coding agent	69.40	46.39	39.30	59.33	61.09	-
	- retry	91.90	83.53	87.70	89.33	85.66	44.22
Problem Solving	- reasoning agent	92.77	54.17	96.90	94.00	85.45	46.36