Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PPoGA: Predictive Plan-on-Graph with Action for Knowledge Graph Question Answering

MinGyu Jeon, SuWan Cho, JaeYoung Shu

TL;DR

The paper tackles the brittleness of KGQA with LLMs caused by fixed high-level plans. It introduces PPoGA, a Planner-Executor architecture augmented with Predictive Processing and a two-level self-correction mechanism that enables both Path Correction and Plan Correction. The method deploys a three-layer Integrated Memory Architecture to manage strategy, step-level dynamics, and knowledge from the KG. Empirical results on GrailQA, CWQ, and WebQSP show state-of-the-art performance among prompting KG-augmented LLM baselines, demonstrating improved robustness and generalization through metacognitive reasoning. This work suggests a promising direction for more autonomous, flexible reasoning systems in complex multi-hop tasks and beyond.

Abstract

Large Language Models (LLMs) augmented with Knowledge Graphs (KGs) have advanced complex question answering, yet they often remain susceptible to failure when their initial high-level reasoning plan is flawed. This limitation, analogous to cognitive functional fixedness, prevents agents from restructuring their approach, leading them to pursue unworkable solutions. To address this, we propose PPoGA (Predictive Plan-on-Graph with Action), a novel KGQA framework inspired by human cognitive control and problem-solving. PPoGA incorporates a Planner-Executor architecture to separate high-level strategy from low-level execution and leverages a Predictive Processing mechanism to anticipate outcomes. The core innovation of our work is a self-correction mechanism that empowers the agent to perform not only Path Correction for local execution errors but also Plan Correction by identifying, discarding, and reformulating the entire plan when it proves ineffective. We conduct extensive experiments on three challenging multi-hop KGQA benchmarks: GrailQA, CWQ, and WebQSP. The results demonstrate that PPoGA achieves state-of-the-art performance, significantly outperforming existing methods. Our work highlights the critical importance of metacognitive abilities like problem restructuring for building more robust and flexible AI reasoning systems.

PPoGA: Predictive Plan-on-Graph with Action for Knowledge Graph Question Answering

TL;DR

The paper tackles the brittleness of KGQA with LLMs caused by fixed high-level plans. It introduces PPoGA, a Planner-Executor architecture augmented with Predictive Processing and a two-level self-correction mechanism that enables both Path Correction and Plan Correction. The method deploys a three-layer Integrated Memory Architecture to manage strategy, step-level dynamics, and knowledge from the KG. Empirical results on GrailQA, CWQ, and WebQSP show state-of-the-art performance among prompting KG-augmented LLM baselines, demonstrating improved robustness and generalization through metacognitive reasoning. This work suggests a promising direction for more autonomous, flexible reasoning systems in complex multi-hop tasks and beyond.

Abstract

Large Language Models (LLMs) augmented with Knowledge Graphs (KGs) have advanced complex question answering, yet they often remain susceptible to failure when their initial high-level reasoning plan is flawed. This limitation, analogous to cognitive functional fixedness, prevents agents from restructuring their approach, leading them to pursue unworkable solutions. To address this, we propose PPoGA (Predictive Plan-on-Graph with Action), a novel KGQA framework inspired by human cognitive control and problem-solving. PPoGA incorporates a Planner-Executor architecture to separate high-level strategy from low-level execution and leverages a Predictive Processing mechanism to anticipate outcomes. The core innovation of our work is a self-correction mechanism that empowers the agent to perform not only Path Correction for local execution errors but also Plan Correction by identifying, discarding, and reformulating the entire plan when it proves ineffective. We conduct extensive experiments on three challenging multi-hop KGQA benchmarks: GrailQA, CWQ, and WebQSP. The results demonstrate that PPoGA achieves state-of-the-art performance, significantly outperforming existing methods. Our work highlights the critical importance of metacognitive abilities like problem restructuring for building more robust and flexible AI reasoning systems.
Paper Structure (8 sections, 1 figure, 2 tables)

This paper contains 8 sections, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. The PPoGA Workflow
  4. Integrated Memory Architecture
  5. Experiment
  6. Experimental Setup
  7. Results
  8. Conclusion

Figures (1)

  • Figure 1: The PPoGA Workflow and Architecture. The framework consists of a Planner (reasoning) and an Executor (action), which interact through a three-layered Integrated Memory Architecture. The workflow follows an iterative cycle of Decomposition, a four-stage Step Cycle (Predict, Act, Observe, Think), and Evaluation, enabling both Plan and Path Correction.