MUSE-AI-Lab/SurveyGraph

Yifan Yuan · Yancheng Zhu · Zuoyu Zhang · Xiaogang Peng† School of Artificial Intelligence, Shenzhen University †Corresponding author EMNLP 2026 Findings English · 简体中文 SurveyGraph: MEKG construction, graph-driven survey generation, and integrated evaluation. TL;DR: LLM survey generators usually retrieve papers as a flat pile of documents, so the result reads like a collage of summaries. A good survey explains how a field evolved — which work extends which, which methods compete, which benchmarks mark turning points. SurveyGraph builds a Methods-Entity Knowledge Graph (MEKG) that records method-level follow-on, competition and shared-resource relations, and uses it as the backbone for core-paper discovery, graph-aware retrieval, hybrid outlining and evidence-gated, evolution-aware writing. Graph-aware retrieval nearly doubles recall over flat lexical baselines (p = 0.004), and 80.6% of generated sentences are fully or partly supported by the papers they cite. Python 3.10–3.12 is tested. Optional components are enabled only when needed: Memgraph as graph…