PolyG: Effective and Efficient GraphRAG with Adaptive Graph Traversal

2025-04-02

AI Summary

Existing GraphRAG approaches employ fixed graph traversal strategies, limiting adaptability to diverse user queries and thereby constraining answer quality and inference efficiency (i.e., latency and token consumption). To address this, we propose the first problem-type–aware adaptive query planning framework for GraphRAG. Our method introduces a fine-grained four-category question taxonomy, a dynamic knowledge graph traversal algorithm tailored to question semantics, and a lightweight planning engine integrated with an LLM-augmented collaborative reasoning module—collectively forming a unified GraphRAG query planner. Extensive experiments demonstrate that our approach achieves a 75% win rate in answer quality over state-of-the-art baselines, accelerates response speed by up to 4×, and significantly reduces token consumption.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

GraphRAG enhances large language models (LLMs) to generate quality answers for user questions by retrieving related facts from external knowledge graphs. Existing GraphRAG methods adopt a fixed graph traversal strategy for fact retrieval but we observe that user questions come in different types and require different graph traversal strategies. As such, existing GraphRAG methods are limited in effectiveness (i.e., quality of the generated answers) and/or efficiency (i.e., response time or the number of used tokens). In this paper, we propose to classify the questions according to a complete four-class taxonomy and adaptively select the appropriate graph traversal strategy for each type of questions. Our system PolyG is essentially a query planner for GraphRAG and can handle diverse questions with an unified interface and execution engine. Compared with SOTA GraphRAG methods, PolyG achieves an overall win rate of 75% on generation quality and a speedup up to 4x on response time.

Problem

• Enhancing LLMs with adaptive graph traversal for diverse questions
• Improving GraphRAG effectiveness and efficiency via question classification
• Unified query planner for varied question types in GraphRAG

Innovation

• Adaptive graph traversal for diverse questions
• Question classification with four-class taxonomy
• Unified interface and execution engine