RGL: A Graph-Centric, Modular Framework for Efficient Retrieval-Augmented Generation on Graphs

TL;DR

Addressing bottlenecks in retrieval-augmented generation on graphs, the paper identifies fixed pipelines and heavy engineering as barriers to scalability. It proposes RGL, a modular library that unifies graph indexing, node and graph retrieval, subgraph construction, tokenization, and generation with a kernel-runtime-API architecture and C++-backed components. The authors report up to $143×$ speedups in graph retrieval and show that dynamic node filtering reduces token consumption while improving downstream task performance. Empirical evaluations on modality completion and abstract generation using real-world graphs such as OGBN-Arxiv demonstrate improved efficiency and generation quality, validating RGL as a scalable platform for RoG applications. Overall, RGL provides a flexible foundation for rapid prototyping and future integration with graph databases and diverse graph formats.

Abstract

Recent advances in graph learning have paved the way for innovative retrieval-augmented generation (RAG) systems that leverage the inherent relational structures in graph data. However, many existing approaches suffer from rigid, fixed settings and significant engineering overhead, limiting their adaptability and scalability. Additionally, the RAG community has largely overlooked the decades of research in the graph database community regarding the efficient retrieval of interesting substructures on large-scale graphs. In this work, we introduce the RAG-on-Graphs Library (RGL), a modular framework that seamlessly integrates the complete RAG pipeline-from efficient graph indexing and dynamic node retrieval to subgraph construction, tokenization, and final generation-into a unified system. RGL addresses key challenges by supporting a variety of graph formats and integrating optimized implementations for essential components, achieving speedups of up to 143x compared to conventional methods. Moreover, its flexible utilities, such as dynamic node filtering, allow for rapid extraction of pertinent subgraphs while reducing token consumption. Our extensive evaluations demonstrate that RGL not only accelerates the prototyping process but also enhances the performance and applicability of graph-based RAG systems across a range of tasks.

Figures (4)

• Figure 1: The pipeline of RAG-on-Graphs.
• Figure 2: Time consumption of different graph retrieval implementations. The learning time involves forward and backward propagation operations, while the retrieval time is introduced by RAG-on-Graph operations.
• Figure 3: An overview of the RGL toolkit.
• Figure 4: Time consumptions (s) vs. query counts across graph retrieval methods and datasets. The light colors denote the original training time, while the dark colors mark the additional time associated with graph retrieval.