EoG: Evolve on Graph

Evolve on Graph: Self-Evolving Knowledge Graph Reasoning without Question Annotations

🏠 Homepage | 💻 GitHub | 🤗 Model

🌟 Overview

Evolve on Graph (EoG) is a self-evolving framework for multi-hop Knowledge Graph Question Answering (KGQA) that improves LLM graph reasoning using only the knowledge graph itself, without human-annotated QA data or question synthesis by closed-source models. Through Graph Self-Play, a single LLM alternates between a Proposer that generates questions and a Solver that answers them through graph exploration. Symbolically Verifiable Topology Synthesis (SVTS) provides reasoning skeletons and correct, complete answer sets by executing first-order logic queries on the graph. A Boundary-Aware Graph Curriculum adapts task difficulty to the Solver's evolving capabilities. EoG consistently outperforms prompting-, retrieval-, and agent-based methods across six benchmarks spanning Freebase, Wikidata, and an industrial knowledge graph.

🛠️ Usage

Model weights will be publicly released upon acceptance of the associated paper. Usage instructions and inference examples will be provided alongside the release. Stay tuned for updates!