Ontology-Driven Data Semantics Discovery for Cyber-Security

We present an architecture for data semantics discovery capable of extracting semantically-rich content from human-readable files without prior specification of the file format. The architecture, based on work at the intersection of knowledge representation and machine learning, includes machine learning modules for automatic file format identification, tokenization, and entity identification. The process is driven by an ontology of domain-specific concepts. The ontology also provides an abstraction layer for querying the extracted data. We provide a general description of the architecture as well as details of the current implementation. Although the architecture can be applied in a variety of domains, we focus on cyber-forensics applications, aiming to allow one to parse data sources, such as log files, for which there are no readily-available parsing and analysis tools, and to aggregate and query data from multiple, diverse systems across large networks. The key contributions of our work are: the development of an architecture that constitutes a substantial step toward solving a highly-practical open problem; the creation of one of the first comprehensive ontologies of cyber assets; the development and demonstration of an innovative, non-trivial combination of declarative knowledge specification and machine learning.