NER Cybersecurity Model v2

Extract security skills, certs, and threats from CVs and job posts in milliseconds.

Model Details

Base Model: en_core_web_sm (spaCy 2.2.5)
Training: Prodigy 1.9.9 with 1805 annotated examples
Accuracy: 99.5% on evaluation set
Version: 2.0 (improved coverage)

Improvements over v1

+24.5% F1 (27.5% → 52.0% on test set)
CVE recognition now working (80% F1, was 0%)
ACRONYM recognition added (66.7% F1, was 0%)
CISO correctly tagged as SECURITY_ROLE (was CERTIFICATION)
805 new training examples covering previously missing types

Entity Types (13)

Category	Entity Types
Core Security	SECURITY_ROLE, SECURITY_TOOL, CERTIFICATION, FRAMEWORK
Threats	CVE, THREAT_TYPE, ATTACK_TECHNIQUE
Skills	TECHNICAL_SKILL, ACRONYM, SECURITY_DOMAIN
Compliance	REGULATION, CONTROL_ID, AUDIT_TERM

Usage

import spacy

nlp = spacy.load("pki/ner-cybersecurity")

doc = nlp("CISO with CISSP. Patched CVE-2024-1234. Expert in SIEM and EDR.")
for ent in doc.ents:
    print(f"{ent.label_}: {ent.text}")

Output:

SECURITY_ROLE: CISO
CERTIFICATION: CISSP
CVE: CVE-2024-1234
SECURITY_TOOL: SIEM
ACRONYM: EDR

Training Data Sources

pki-ad-match role YAMLs (62 roles)
agent-nexus-85 agent presets (90 agents)
Synthetic examples with cybersecurity entities
805 new examples for underrepresented types

Requirements

spaCy 2.2.x
Python 3.6-3.8

License

MIT