This study presents an Adaptive Backbone Extraction Framework designed to address the challenges of network heterogeneity. Unlike traditional methods that apply a uniform approach to the entire network, this framework distinguishes between local, densely connected components and their broader interconnections, termed global components. The framework captures fine-grained and overarching network structures, allowing customized backbone extraction techniques within each component. Experiments on realworld networks demonstrate that this approach preserves critical local and global properties more effectively than classical methods, making it a versatile tool for diverse network applications.