A classifier trained on 1,207 expert-mapped CVEs roughly doubles recall@5 over a zero-shot baseline, while LLM-generated labels at ~0.39 expert agreement provide no reliable gain and degrade rare-technique coverage at ~1,000 added CVEs.
vulnerability-attack-technique-classification- roberta-base (Revision 3b98a1d)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion
A classifier trained on 1,207 expert-mapped CVEs roughly doubles recall@5 over a zero-shot baseline, while LLM-generated labels at ~0.39 expert agreement provide no reliable gain and degrade rare-technique coverage at ~1,000 added CVEs.