PlanQA: A plan-execute-reason framework for knowledge graph question answering using large language models

doi:10.1016/j.eswa.2026.132215

PlanQA: A plan-execute-reason framework for knowledge graph question answering using large language models

2026-03-26

0

PRE

AI

DOI：10.1016/j.eswa.2026.132215

Original

Original request for help

Save

Abstract

En 中文

Recently, Large Language Models (LLMs) have demonstrated remarkable performance in many tasks. However, since relying solely on the information stored in LLMs brings significant risks, recent works have focused on utilizing external information sources such as Knowledge Graphs along with LLMs. In this paper, we focus on using LLMs for answering questions based on the information stored in Knowledge Graphs, which is known as Knowledge Graph Question Answering (KGQA). Previous KGQA methods suffer from two limitations which make it difficult to reliably retrieve the information required for answering questions. They process entities that are connected to topic entities with the same relation path separately, and do not precisely define what steps can be used for answering a question. To address these issues, we propose a new Plan-Execute-Reason framework for KGQA called PlanQA. Specifically, we introduce a versatile and reliable planning method, which allows LLMs to reliably solve questions of diverse types. We also propose an algorithm for exploring diverse relation paths called Execution Alteration. PlanQA outperforms existing state-of-the-art models on four public KGQA datasets: WebQSP, CWQ, GrailQA, and GraphQA.

Keywords:

PlanQA

Knowledge Graph Question Answering

Large Language Models

Plan-Execute-Reason framework

Execution Alteration

PlanQA: A plan-execute-reason framework for knowledge graph question answering using large language models

Abstract

Journal

Organization

Cited Papers