Extract Class refactoring is essential for decomposing large, complex classes to improve code maintainability and readability. Traditional refactoring tools rely heavily on metrics like cohesion and coupling, which often require expert judgment and are not universally applicable. This research proposes a novel approach leveraging deep class property graphs and advanced graph neural networks to automate the identification of refactoring opportunities. By integrating deep semantic properties and fine-grained structural dependencies, this method aims to reduce reliance on expert knowledge and improve the precision and adaptability of refactoring suggestions. Future work will explore hypergraph learning to capture more complex code relationships, further enhancing the proposed method’s effectiveness.