LiteRAG speeds up graph retrieval for multi-hop question answering
LiteRAG: Cost-Efficient Graph-Based Retrieval-Augmented Generation
Information RetrievalArtificial IntelligenceComputation and Language
Summary
Answering complex questions that need multiple steps often requires looking through lots of information, which can be slow and costly. The authors develop LiteRAG, a new way to quickly find and organize relevant information from a network of documents without needing costly language model control during retrieval. Their method is much faster and uses fewer words in the final context while maintaining or improving answer quality. LiteRAG’s smart shortcuts and ways to avoid busy network hubs help it work efficiently.
What this means in practice
- •For software engineering teams: Build systems that answer distributed systems questions faster by using efficient graph-based retrieval that reduces computation during query time.
- •For enterprise ai developers: Create cost-effective AI tools that generate responses grounded in relevant technical documents while minimizing token usage and response delay.$Commercial implications: Enables AI applications providing efficient, accurate knowledge retrieval and answer generation to enterprise customers needing cost control.
Authors
Daniel Alejandro Coll Tejeda, Pedro García López, Daniel Barcelona-Pons
Abstract
Graph-based retrieval can improve multi-hop question answering, but existing approaches often incur high query-time costs and produce diffuse, oversized contexts that reduce generation efficiency. We present LiteRAG, a graph-based retrieval method that replaces expensive retrieval-time LLM control with query-conditioned algorithmic exploration and reasoning-chain context construction. On DistComp, a benchmark for multi-hop retrieval over distributed-systems papers, LiteRAG attains the highest overall quality among the evaluated methods (0.798) while reducing per-query latency by over 100$\times$ and cost by over 99% relative to GraphRAG Global and DRIFT. On UltraDomain, it matches LinearRAG on overall quality while using about 14$\times$ fewer tokens. An ablation study indicates that LiteRAG's query-adaptive thresholding and community-aware hub penalization are the main drivers of its token-efficiency gains.