SurveyG: A Multi-Agent LLM Framework with Hierarchical Citation Graph for Automated Survey Generation
TLDR
SurveyG uses hierarchical citation graphs and multi-agent LLMs to generate coherent, structured survey papers with human and LLM evaluation.
Reasoning
The paper introduces a novel framework that leverages hierarchical citation graphs to capture research evolution, addressing a key limitation of existing survey generation methods. Strengths include a structured multi-agent validation stage and empirical evaluation with human experts. Weaknesses are not apparent from the abstract alone, but the reliance on LLM-as-a-judge may introduce bias.
Read-first score
Read-first score 58.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 61.
Field roles
Rank sensitivity
Stability: volatile; rank range: 24.
Keyword Scores
Deep Analysis
Innovations
- Hierarchical citation graph with three layers (Foundation, Development, Frontier) capturing research evolution
- Multi-agent framework combining horizontal search within layers and vertical depth traversal across layers for multi-level summaries
- Multi-agent validation stage ensuring consistency, coverage, and factual accuracy
- Integration of citation dependencies and semantic relatedness into survey generation
Methodology
SurveyG constructs a hierarchical citation graph where nodes are papers and edges represent citation and semantic relations. Papers are organized into Foundation, Development, and Frontier layers; the agent performs horizontal and vertical traversal to generate multi-level summaries, which are consolidated into a structured outline. A multi-agent validation stage then refines the final survey.
Key Results
SurveyG outperforms state-of-the-art frameworks in human expert and LLM-as-a-judge evaluations, producing surveys that are more comprehensive and better structured to the field's knowledge taxonomy.