Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

KGFR: A Foundation Retriever for Generalized Knowledge Graph Question Answering

Yuanning Cui, Zequn Sun, Wei Hu, Zhangjie Fu

TL;DR

The paper addresses knowledge-grounded question answering by combining large language models with a structured, scalable KG retriever. It introduces KGFR, a zero-shot, non-finetuned retriever that uses LLM-generated unified relation descriptions and question-conditioned propagation, aided by Asymmetric Progressive Propagation to scale to million-scale graphs. A controller–retriever loop enables iterative reasoning with multi-level retrieval (node, edge, path) and a reflection mechanism, enabling robust generalization to unseen KGs and datasets. Experiments across seven KGQA benchmarks show strong accuracy, scalability, and transferability, with ablation studies validating the contribution of each component and evidence that the approach reduces reliance on LLM finetuning while maintaining interpretability and efficiency.

Abstract

Large language models (LLMs) excel at reasoning but struggle with knowledge-intensive questions due to limited context and parametric knowledge. However, existing methods that rely on finetuned LLMs or GNN retrievers are limited by dataset-specific tuning and scalability on large or unseen graphs. We propose the LLM-KGFR collaborative framework, where an LLM works with a structured retriever, the Knowledge Graph Foundation Retriever (KGFR). KGFR encodes relations using LLM-generated descriptions and initializes entities based on their roles in the question, enabling zero-shot generalization to unseen KGs. To handle large graphs efficiently, it employs Asymmetric Progressive Propagation (APP)- a stepwise expansion that selectively limits high-degree nodes while retaining informative paths. Through node-, edge-, and path-level interfaces, the LLM iteratively requests candidate answers, supporting facts, and reasoning paths, forming a controllable reasoning loop. Experiments demonstrate that LLM-KGFR achieves strong performance while maintaining scalability and generalization, providing a practical solution for KG-augmented reasoning.

KGFR: A Foundation Retriever for Generalized Knowledge Graph Question Answering

TL;DR

The paper addresses knowledge-grounded question answering by combining large language models with a structured, scalable KG retriever. It introduces KGFR, a zero-shot, non-finetuned retriever that uses LLM-generated unified relation descriptions and question-conditioned propagation, aided by Asymmetric Progressive Propagation to scale to million-scale graphs. A controller–retriever loop enables iterative reasoning with multi-level retrieval (node, edge, path) and a reflection mechanism, enabling robust generalization to unseen KGs and datasets. Experiments across seven KGQA benchmarks show strong accuracy, scalability, and transferability, with ablation studies validating the contribution of each component and evidence that the approach reduces reliance on LLM finetuning while maintaining interpretability and efficiency.

Abstract

Large language models (LLMs) excel at reasoning but struggle with knowledge-intensive questions due to limited context and parametric knowledge. However, existing methods that rely on finetuned LLMs or GNN retrievers are limited by dataset-specific tuning and scalability on large or unseen graphs. We propose the LLM-KGFR collaborative framework, where an LLM works with a structured retriever, the Knowledge Graph Foundation Retriever (KGFR). KGFR encodes relations using LLM-generated descriptions and initializes entities based on their roles in the question, enabling zero-shot generalization to unseen KGs. To handle large graphs efficiently, it employs Asymmetric Progressive Propagation (APP)- a stepwise expansion that selectively limits high-degree nodes while retaining informative paths. Through node-, edge-, and path-level interfaces, the LLM iteratively requests candidate answers, supporting facts, and reasoning paths, forming a controllable reasoning loop. Experiments demonstrate that LLM-KGFR achieves strong performance while maintaining scalability and generalization, providing a practical solution for KG-augmented reasoning.

Paper Structure

This paper contains 39 sections, 12 equations, 4 figures, 13 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLM reasoning with KGs
  4. KG foundation models
  5. Preliminaries
  6. Knowledge graph question answering
  7. Generalized KGQA
  8. Knowledge Graph Foundation Retriever
  9. Relation Initialization via Unified Descriptions
  10. Question-Conditioned Message-Passing
  11. Asymmetric Progressive Propagation
  12. Pre-training Objective
  13. Collaborative Question Answering with LLMs
  14. KGFR-based Knowledge Retrieval
  15. LLM-based Generation and Reflection
  16. ...and 24 more sections

Figures (4)

  • Figure 1: Framework of LLM-KGFR. Given a question and a KG, we encode the question and relations using BERT and perform asymmetric progressive propagation from the topic entity. The KGFR then conducts multi-level (node, edge, path) retrieval to iteratively generate and refine answers with the LLM until the final answer is confirmed.
  • Figure 2: Illustrative prompt used for generating unified textual descriptions of relations.
  • Figure 3: KGFR only top-$k$ retrieval results between WebQSP and CWQ
  • Figure 4: Acc. of multiple-choice QA