Complex Querying of Cyber Knowledge Graph for Robust Cyber Defense
summary
A production Neo4j knowledge graph ingests CVE/CWE/CAPEC/CPE feeds (336,000 CVEs, ~620,000 CPEs) and adds ~350,000 inferred edges using V2W BERT and VWC BERT; a GraphRAG front end auto-generates Cypher, retrieves evidence, and returns answers for 600 benchmark questions across six categories. Semantic search uses Gemini embeddings (≈3,000 dims) and 1,024-d Query2Box vectors with TransE/box reasoning; semantic nearest-neighbor yields ~97.5% CWE agreement but some intel intersection queries show low completeness, while live updating and GraphRAG integration proof operational value.
AI-written summary · about
chapters (11)
- 0:00Introduction and team overview
- 1:56Cybersecurity knowledge graph goals
- 2:58Knowledge graph data sources
- 6:27Graph construction and automation
- 8:57Path-based queries in knowledge graph
- 11:51Handling missing links in graph
- 13:39Semantic information and GraphRAG
- 17:56Vector embeddings for CVEs
- 21:47Advanced embedding-based querying
- 25:39Operational integration by Joe
- 27:50Q&A and future directions
related talks
26mCyRAG: Retrieval Augmented Generation for Robust Cyber Defense
15mGraph-Driven Knowledge Retrieval: Neo4j for Healthcare Building Codes
24mYour company, as a Knowledge Graph - the foundation of cybersecurity’s future
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
33mSmarter Industrial Fault Diagnosis with Knowledge Graphs and Large Language Models
38mLeveraging Knowledge Graphs for RAG: A Smarter Approach to Contextual AI Applications
30mAgentic RAG with Knowledge Graphs Explained
01h 00mGraph Mastery to AI