Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Iterative Question-Guided Framework for Knowledge Base Question Answering

Shuai Wang, Yinan Yu

TL;DR

This work addresses factuality and reasoning challenges in knowledge-intensive QA by integrating knowledge graphs with large language models. It introduces iQUEST, an iterative, question-guided KBQA framework that decomposes complex queries into sub-questions via IQG-LLM and augments reasoning with a Graph Neural Network that looks ahead through 2-hop neighbors. The approach comprises three modules—Iterative Question Guidance, Two-hop Entity Exploration, and Answer Extraction—delivering significant improvements across CWQ, WebQSP, WebQuestion, and GrailQA without LLM fine-tuning. Extensive ablations show the contributions of KG augmentation, iterative sub-question generation, and 2-hop look-ahead, underscoring robustness and generalization across diverse datasets and LLMs. The results suggest iQUEST can enhance reliable, explainable KBQA in real-world, knowledge-intensive applications, albeit with runtime overhead and KG completeness as future focal points.

Abstract

Large Language Models (LLMs) excel in many natural language processing tasks but often exhibit factual inconsistencies in knowledge-intensive settings. Integrating external knowledge resources, particularly knowledge graphs (KGs), provides a transparent and updatable foundation for more reliable reasoning. Knowledge Base Question Answering (KBQA), which queries and reasons over KGs, is central to this effort, especially for complex, multi-hop queries. However, multi-hop reasoning poses two key challenges: (1)~maintaining coherent reasoning paths, and (2)~avoiding prematurely discarding critical multi-hop connections. To tackle these challenges, we introduce iQUEST, a question-guided KBQA framework that iteratively decomposes complex queries into simpler sub-questions, ensuring a structured and focused reasoning trajectory. Additionally, we integrate a Graph Neural Network (GNN) to look ahead and incorporate 2-hop neighbor information at each reasoning step. This dual approach strengthens the reasoning process, enabling the model to explore viable paths more effectively. Detailed experiments demonstrate the consistent improvement delivered by iQUEST across four benchmark datasets and four LLMs.

An Iterative Question-Guided Framework for Knowledge Base Question Answering

TL;DR

This work addresses factuality and reasoning challenges in knowledge-intensive QA by integrating knowledge graphs with large language models. It introduces iQUEST, an iterative, question-guided KBQA framework that decomposes complex queries into sub-questions via IQG-LLM and augments reasoning with a Graph Neural Network that looks ahead through 2-hop neighbors. The approach comprises three modules—Iterative Question Guidance, Two-hop Entity Exploration, and Answer Extraction—delivering significant improvements across CWQ, WebQSP, WebQuestion, and GrailQA without LLM fine-tuning. Extensive ablations show the contributions of KG augmentation, iterative sub-question generation, and 2-hop look-ahead, underscoring robustness and generalization across diverse datasets and LLMs. The results suggest iQUEST can enhance reliable, explainable KBQA in real-world, knowledge-intensive applications, albeit with runtime overhead and KG completeness as future focal points.

Abstract

Large Language Models (LLMs) excel in many natural language processing tasks but often exhibit factual inconsistencies in knowledge-intensive settings. Integrating external knowledge resources, particularly knowledge graphs (KGs), provides a transparent and updatable foundation for more reliable reasoning. Knowledge Base Question Answering (KBQA), which queries and reasons over KGs, is central to this effort, especially for complex, multi-hop queries. However, multi-hop reasoning poses two key challenges: (1)~maintaining coherent reasoning paths, and (2)~avoiding prematurely discarding critical multi-hop connections. To tackle these challenges, we introduce iQUEST, a question-guided KBQA framework that iteratively decomposes complex queries into simpler sub-questions, ensuring a structured and focused reasoning trajectory. Additionally, we integrate a Graph Neural Network (GNN) to look ahead and incorporate 2-hop neighbor information at each reasoning step. This dual approach strengthens the reasoning process, enabling the model to explore viable paths more effectively. Detailed experiments demonstrate the consistent improvement delivered by iQUEST across four benchmark datasets and four LLMs.

Paper Structure

This paper contains 30 sections, 5 equations, 2 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. KBQA
  4. Question Decomposition
  5. Entity Exploration
  6. Problem defination
  7. Method
  8. Iterative Question Guidance
  9. Two-Hop Based Entity Exploration
  10. Answer Extraction LLM (AE-LLM)
  11. Experiments
  12. Datasets
  13. Main results
  14. Ablation Study
  15. Assumptions
  16. ...and 15 more sections

Figures (2)

  • Figure 1: Examples of multi-step reasoning in a knowledge graph, each starting from a topic entity. Question 1 illustrates how maintaining coherent reasoning paths can be challenging, whereas Question 2 shows how critical multi-hop connections may be prematurely discarded. Entities marked with ✓ are selected in the reasoning path, and those marked with × are unselected.
  • Figure 2: Overview of our framework, illustrating the reasoning trajectory, workflow, and multi-step reasoning process over KGs.