HiGraph: A Large-Scale Hierarchical Graph Dataset

Graph-based methods have shown great promise in malware analysis, yet the lack of large-scale, hierarchical graph datasets limits further advances in this field. To bridge this gap, we introduceHIGRAPH, a novel, large-scale dataset that models each application as a hierarchical graph: a local Control Flow Graph (CFG) capturing intra-function logic and a global Function Call Graph (FCG) capturing inter-function interactions.

This hierarchical design facilitates the development of robust detection models that are more resilient to obfuscation, model aging, and malware evolution. HIGRAPH contains over 200M control flow graphs and 595K function call graphs, preserving rich semantic and structural information crucial for analyzing sophisticated malware behaviors. We provide an in-depth analysis of HIGRAPH and highlight its potential as a benchmark dataset for advancing hierarchical graph learning in cybersecurity.