Ontology matching addresses data heterogeneity challenges while enhancing system interoperability, serving as a critical mechanism for knowledge fusion. However, most existing approaches rely on predefined vocabularies and logic rules, limiting their ability to capture deeper semantic information about entities. Recent advancements in deep learning and Large Language Models (LLMs) have positively contributed to improving ontology matching. Based on this, we propose LGOM, a novel ontology matching method that integrates LLMs with graph structure learning network. Specifically, we leverage LLMs to generate high-quality textual descriptions of entities, enriching their semantic representations. Additionally, we introduce a graph structure learning network that aggregates information from neighboring entities within the ontology, enabling more comprehensive and discriminative representations for alignment. Experimental results demonstrate that LGOM outperforms baseline models and highlights the importance of incorporating both entity descriptions and structural information in ontology matching.