Guided Navigation in Knowledge-Dense Environments: Structured Semantic Exploration with Guidance Graphs

Dehao Tao, Guangjie Liu, weizheng, Yongfeng Huang, Minghu Jiang

Keywords: knowledge graph, LLM, guidance graph, question answering

Abstract:

While Large Language Models (LLMs) exhibit strong linguistic capabilities, their reliance on static knowledge and opaque reasoning processes limits their performance in knowledge-intensive tasks. Although Knowledge Graphs (KGs) can mitigate this, existing retrieval methods are caught in a fundamental granularity trap: query-guided search leads to wasteful redundancy, while clue-guided traversal struggles with contextual reasoning in multi-hop scenarios.
To address these limitations, we propose Guidance-Graph-guided Knowledge Exploration (GG-Explore), a novel framework that introduces an intermediate Guidance Graph to bridge unstructured queries and structured knowledge retrieval. This Guidance Graph acts as a lightweight semantic blueprint, abstracting the structure of potential answers to constrain the search space without sacrificing contextual breadth.
Leveraging this graph, GG-Explore employs a hybrid pruning strategy, combining a model-free Structural Alignment that filters candidates using graph constraints with a Context-Aware Semantic Alignment module that refines the results by enforcing semantic consistency.
Extensive experiments show our method achieves superior efficiency and outperforms SOTA, especially on complex tasks, while maintaining strong performance with smaller LLMs, demonstrating practical value.

Primary Area: interpretability and explainable AI

Submission Number: 23056