Knowledge Retrieval in LLM Gaming: A Shift from Entity-Centric to Goal-Oriented Graphs

2025-05-24

AI Summary

Large language models (LLMs) exhibit limited stepwise reasoning capability in complex, multi-step tasks such as game playing. Existing retrieval-augmented methods—e.g., GraphRAG—rely on entity-centric knowledge graphs with fragmented structures and overly dense local connections, hindering coherent long-range reasoning. To address this, we propose Goal-oriented Graphs (GoG), a novel framework that models reasoning as a top-down, recursive decomposition of high-level goals into subgoals, where nodes represent goals and directed edges encode logical dependencies. GoG shifts knowledge graph modeling from entity-centric to goal-centric paradigms, enabling explicit retrieval and structured orchestration of reasoning paths. The framework comprises four components: goal graph construction, hierarchical goal retrieval, logical edge encoding, and GoG-enhanced prompt engineering. Experiments on the Minecraft platform demonstrate that GoG significantly outperforms GraphRAG and other baselines, validating the efficacy of goal-chain reasoning for enhancing LLMs’ long-horizon reasoning capabilities.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Large Language Models (LLMs) demonstrate impressive general capabilities but often struggle with step-by-step reasoning, especially in complex applications such as games. While retrieval-augmented methods like GraphRAG attempt to bridge this gap through cross-document extraction and indexing, their fragmented entity-relation graphs and overly dense local connectivity hinder the construction of coherent reasoning. In this paper, we propose a novel framework based on Goal-Oriented Graphs (GoGs), where each node represents a goal and its associated attributes, and edges encode logical dependencies between goals. This structure enables explicit retrieval of reasoning paths by first identifying high-level goals and recursively retrieving their subgoals, forming coherent reasoning chains to guide LLM prompting. Our method significantly enhances the reasoning ability of LLMs in game-playing tasks, as demonstrated by extensive experiments on the Minecraft testbed, outperforming GraphRAG and other baselines.

Problem

• Enhance LLM step-by-step reasoning in complex games
• Improve fragmented entity-relation graphs in retrieval methods
• Construct coherent reasoning chains using Goal-Oriented Graphs

Innovation

• Goal-Oriented Graphs structure for coherent reasoning
• Recursive retrieval of subgoals to form reasoning chains
• Enhanced LLM reasoning in gaming tasks