TopoCorr-KG
Topology-Aware Retrieval and Correlation Across Disparate Cybersecurity Knowledge Graphs
A Florida State University research project funded by the ReliaQuest Innovation Challenge. TopoCorr-KG retrieves and correlates entities across multiple authoritative cybersecurity knowledge graphs without physical integration, returning ranked, source-attributed evidence paths for analyst verification.
Project Overview
Authoritative cybersecurity knowledge is distributed across several disjoint catalogs maintained by MITRE and NIST — including ATT&CK (attacker behaviors), CAPEC (attack patterns), CWE (software-weakness categories), and CVE/NVD (disclosed vulnerabilities). Centralizing these sources is often impractical due to engineering overhead, consistency burdens, and privacy or contractual constraints — leaving analysts to manually correlate entities across silos.
TopoCorr-KG queries each source independently and composes locally relevant subgraphs into ranked evidence paths:
Behavior → Attack Pattern → Software Weakness → Vulnerability
Each step carries full provenance, so analysts can verify or reject candidate paths in seconds. The AI component is a learned cross-source correlation model that combines semantic embeddings, graph-structural descriptors, and supervised link-prediction signals.
Approach
1. Per-Source Graphs
Each MITRE/NIST catalog is loaded into its own logical Neo4j database. The four databases are kept separate by design.
2. Hybrid Search Index
Each source receives a hybrid index combining classical keyword retrieval with a dense sentence-transformer encoder, plus cached structural and topological descriptors.
3. Query-Time Retrieval
Each index is searched independently. Local subgraph context is fetched on demand — no global scan or merge required.
4. Cross-Source Ranking
Candidates are linked into evidence paths scored by textual similarity, structural similarity, and a learned cross-source link-prediction signal trained on official MITRE mappings.
Project Team
Astrit Tola
Principal Investigator
Postdoctoral Scholar, Department of Mathematics, FSU. Research focuses on topological data analysis, graph machine learning, and AI for science and security. Leads TopoML&AI @ FSU.
Malbor Asllani
Co-Principal Investigator
Assistant Professor of Interdisciplinary Mathematics at FSU. Brings complementary expertise in network science and dynamical processes on graphs.
Ece Karacam
Graduate Research Assistant
Bio coming soon.
Sebastian Powers
Undergraduate Research Assistant
Bio coming soon.
Funding
ReliaQuest Innovation Challenge — Florida State University
This project is supported by the ReliaQuest Innovation Challenge at Florida State University. We thank ReliaQuest and FSU for their support of research at the intersection of AI, topology, and cybersecurity.
Contact
For questions, collaboration opportunities, or inquiries, please contact the PI: