KG-Infused RAG: Augmenting Corpus-Based RAG with External Knowledge Graphs
Dingjun Wu, Yukun Yan, Zhenghao Liu, Zhiyuan Liu, Maosong Sun
TL;DR
KG-Infused RAG enhances retrieval-augmented generation by integrating large-scale external knowledge graphs with spreading activation, improving factual accuracy and interpretability across multiple QA benchmarks.
Contribution
This work introduces a novel framework that incorporates pre-existing knowledge graphs into RAG, using spreading activation to improve retrieval and generation with lower cost and higher performance.
Findings
Outperforms vanilla RAG by 3.9% to 17.8% on five QA benchmarks
Achieves better results than KG-based methods like GraphRAG and LightRAG
Enhances existing RAG models when integrated with Self-RAG and DeepNote
Abstract
Retrieval-Augmented Generation (RAG) improves factual accuracy by grounding responses in external knowledge. However, existing RAG methods either rely solely on text corpora and neglect structural knowledge, or build ad-hoc knowledge graphs (KGs) at high cost and low reliability. To address these issues, we propose KG-Infused RAG, a framework that incorporates pre-existing large-scale KGs into RAG and applies spreading activation to enhance both retrieval and generation. KG-Infused RAG directly performs spreading activation over external KGs to retrieve relevant structured knowledge, which is then used to expand queries and integrated with corpus passages, enabling interpretable and semantically grounded multi-source retrieval. We further improve KG-Infused RAG through preference learning on sampled key stages of the pipeline. Experiments on five QA benchmarks show that KG-Infused RAG…
Peer Reviews
Decision·Submitted to ICLR 2026
* The pipeline is clearly articulated and easy to follow. * The proposed method is evaluated using a variety of LLMs. * The paper provides comprehensive experimental results, including a clear ablation study that highlights the contributions of different model components.
* The novelty appears somewhat incremental, as previous research has also explored the use of KGs for QA and retrieval enhancement. The authors should more strongly emphasis what is new vs what is incremental. In particular, since the paper claims novelty in leveraging a pre-existing KG to enhance RAG, it should include comparisons with existing methods that also utilize pre-existing KGs (e.g., [1]). Additionally, it would be valuable to compare with the extensive body of work on QA systems that
This paper addresses a key gap between corpus-based and KG-based RAG approaches, namely, how to combine structured and unstructured knowledge for retrieval and reasoning efficiently. Demonstrates gains on multiple QA datasets and provides comparisons with both corpus-only and KG-constructed baselines. The framework can potentially be applied to enhance other RAG systems.
1. The writing of the paper is not very clear: The spreading-activation process (e.g., weighting, termination, or entity-selection heuristics) is only partially described. The use of spreading activation is intuitively motivated but lacks rigorous analysis or ablation exploring why it outperforms simpler entity expansion methods. 2. All benchmarks are QA-centric; there is no evidence the approach generalizes to other tasks (e.g., reasoning, summarization, scientific retrieval). Also, ther paper
S1. Paper very well written, with clear explanations and formalisms. S2. The method is comprehensive, covering various aspects of RAG. S3. Good experimental study showing quality improvements.
W1. The paper is not comprehensive in comparison w. sota solutions regarding RAG on KGs: [1] KERAG: Knowledge-Enhanced Retrieval-Augmented Generation for Advanced Question Answering [2] Think-on-Graph: Deep and Responsible Reasoning of Large Language Model on Knowledge Graph Section 3.3 is very similar to [1], reducing novelty of the work. W2. Using KG for query extension can be interesting. However, the retrieved KG subgraph can be huge and the query extension can thus bring in a lot of retr
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAdvanced Graph Neural Networks · Topic Modeling · Information Retrieval and Search Behavior