Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CoG: Controllable Graph Reasoning via Relational Blueprints and Failure-Aware Refinement over Knowledge Graphs

Yuanxiang Liu, Songze Li, Xiaoke Guo, Zhaoyan Gong, Qifei Zhang, Huajun Chen, Wen Zhang

TL;DR

CoG tackles reliability and grounding issues in LLM-based KG reasoning by introducing a training-free, dual-process-inspired framework that jointly uses offline relational blueprints for stable guidance and online failure-aware refinement for robustness. The method constructs a compact library of relational blueprint templates from training data and adapts them online as soft structural priors to constrain multi-hop KG exploration, while a diagnostic backtracking module remedies dead-ends and ensures verifiability. Across WebQSP, CWQ, and GrailQA, CoG consistently surpasses state-of-the-art baselines in accuracy and efficiency, including zero-shot performance on unseen query structures and robustness to schema changes. The approach offers a practical, scalable path to KG-grounded reasoning without costly fine-tuning, highlighting the value of explicit structural priors coupled with dynamic self-correction for complex KGQA tasks.

Abstract

Large Language Models (LLMs) have demonstrated remarkable reasoning capabilities but often grapple with reliability challenges like hallucinations. While Knowledge Graphs (KGs) offer explicit grounding, existing paradigms of KG-augmented LLMs typically exhibit cognitive rigidity--applying homogeneous search strategies that render them vulnerable to instability under neighborhood noise and structural misalignment leading to reasoning stagnation. To address these challenges, we propose CoG, a training-free framework inspired by Dual-Process Theory that mimics the interplay between intuition and deliberation. First, functioning as the fast, intuitive process, the Relational Blueprint Guidance module leverages relational blueprints as interpretable soft structural constraints to rapidly stabilize the search direction against noise. Second, functioning as the prudent, analytical process, the Failure-Aware Refinement module intervenes upon encountering reasoning impasses. It triggers evidence-conditioned reflection and executes controlled backtracking to overcome reasoning stagnation. Experimental results on three benchmarks demonstrate that CoG significantly outperforms state-of-the-art approaches in both accuracy and efficiency.

CoG: Controllable Graph Reasoning via Relational Blueprints and Failure-Aware Refinement over Knowledge Graphs

TL;DR

CoG tackles reliability and grounding issues in LLM-based KG reasoning by introducing a training-free, dual-process-inspired framework that jointly uses offline relational blueprints for stable guidance and online failure-aware refinement for robustness. The method constructs a compact library of relational blueprint templates from training data and adapts them online as soft structural priors to constrain multi-hop KG exploration, while a diagnostic backtracking module remedies dead-ends and ensures verifiability. Across WebQSP, CWQ, and GrailQA, CoG consistently surpasses state-of-the-art baselines in accuracy and efficiency, including zero-shot performance on unseen query structures and robustness to schema changes. The approach offers a practical, scalable path to KG-grounded reasoning without costly fine-tuning, highlighting the value of explicit structural priors coupled with dynamic self-correction for complex KGQA tasks.

Abstract

Large Language Models (LLMs) have demonstrated remarkable reasoning capabilities but often grapple with reliability challenges like hallucinations. While Knowledge Graphs (KGs) offer explicit grounding, existing paradigms of KG-augmented LLMs typically exhibit cognitive rigidity--applying homogeneous search strategies that render them vulnerable to instability under neighborhood noise and structural misalignment leading to reasoning stagnation. To address these challenges, we propose CoG, a training-free framework inspired by Dual-Process Theory that mimics the interplay between intuition and deliberation. First, functioning as the fast, intuitive process, the Relational Blueprint Guidance module leverages relational blueprints as interpretable soft structural constraints to rapidly stabilize the search direction against noise. Second, functioning as the prudent, analytical process, the Failure-Aware Refinement module intervenes upon encountering reasoning impasses. It triggers evidence-conditioned reflection and executes controlled backtracking to overcome reasoning stagnation. Experimental results on three benchmarks demonstrate that CoG significantly outperforms state-of-the-art approaches in both accuracy and efficiency.
Paper Structure (62 sections, 6 equations, 8 figures, 9 tables)

This paper contains 62 sections, 6 equations, 8 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Methodology
  4. Offline Relational Blueprint Construction
  5. Online Blueprint-Guided KG Exploration
  6. Initialization and Structural Planning
  7. Blueprint-Guided KG Exploration
  8. Failure-Aware Refinement
  9. Experiments
  10. Experiment Setup
  11. Main Results
  12. Ablation Study
  13. Efficiency Analysis
  14. Related Work
  15. Conclusion
  16. ...and 47 more sections

Figures (8)

  • Figure 1: Illustration of cognitive rigidity in existing works and how CoG addresses it: (I) Error cascading from indiscriminate exploration (top), mitigated by Relational Blueprints for blueprint-guided relation reranking and pruning (a vs. b); (II) structural misalignment from myopic decisions (bottom), corrected by Failure-Aware Refinement via backtracking and controlled fallback (c vs. d).
  • Figure 2: Overview of the CoG framework. CoG instantiates Dual-Process Theory as a cooperative reasoning loop. Left: Offline blueprint construction abstracts relational sequences from training paths into a searchable template library. Middle (System 1): Online blueprint-guided exploration adapts query-specific blueprints as soft structural constraints to rerank and prune candidate relations. Right (System 2): Failure-aware refinement monitors the reasoning process; upon failure signals (e.g., search stagnation), it performs evidence-conditioned reflection and targeted backtracking to recover a verifiable answer (e.g., enforcing the temporal constraint "last won").
  • Figure 3: Sensitivity analysis of reranking weights. The blue line tracks performance as $\lambda_{loc}$ varies. The orange triangles represent variants with suboptimal internal structural ratios at the peak $\lambda_{loc}=0.6$. The grey square denotes the naive uniform baseline. The explicit weight configuration $(\lambda_{loc}, \lambda_{step}, \lambda_{glob})$ is annotated for each point.
  • Figure 4: Performance breakdown by query type. Our method achieves significant accuracy gains, particularly in structurally complex categories like Conjunction (+4.7%) and Superlative (+4.6%).
  • Figure 5: Outcome distribution of the refinement process. The "Right" segment represents the proportion of queries successfully rectified by the module after being triggered.
  • ...and 3 more figures