Abstract
Large Language Models (LLMs) possess impressive reasoning abilities but areprone to generating incorrect information, often referred to as hallucinations.While incorporating external Knowledge Graphs (KGs) can partially mitigate thisissue, existing methods primarily treat KGs as static knowledge repositories,overlooking the critical disparity between KG and LLM knowledge, and failing tofully exploit the reasoning capabilities inherent in KGs. To address theselimitations, we propose Pyramid-Driven Alignment (PDA), a novel framework forseamlessly integrating LLMs with KGs. PDA utilizes Pyramid Principle analysisto construct a hierarchical pyramid structure. This structure is designed toreflect the input question and generate more validated deductive knowledge,thereby enhancing the alignment of LLMs and KGs and ensuring more cohesiveintegration. Furthermore, PDA employs a recursive mechanism to harness theunderlying reasoning abilities of KGs, resulting in more accurate knowledgeretrieval for question-answering tasks. Our experimental results reveal asubstantial performance advantage of PDA over state-of-the-art baselines, withimprovements reaching 26.70% and 26.78%.