Skip to content
Conference Open access

HCE-ATT&CK: Release-Valid and Evidence-Calibrated Mapping of Cyber Threat Intelligence to MITRE ATT&CK

Sep 2026 · 2026 10th International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) · pp. 1255-1262 · 0 citations · 23 references

Abstract

Cyber threat intelligence (CTI) reports describe adversary behavior in prose, and analysts still map that prose to MITRE ATT&CK techniques largely by hand. Automatic mappers exist, but their published scores are hard to compare, because ATT&CK releases, dataset views, document splits, and invalid-label handling all differ between papers. We rebuild one evaluation contract on AnnoCTR: all 400 linking reports audited, the upstream 70/16/34 document split recovered and hash-checked, one canonical w_con annotation view per report, and every candidate validated against Enterprise ATT&CK v19.1. On this contract we train HCE-ATT&CK, a re-ranker that keeps the relevance prior of a pretrained cybersecurity cross-encoder, learns a zero-initialized hierarchy-aware residual over release-valid candidate cards, and scores confidence with a sufficiency head separate from the ranking path. On the 917-example test split it reaches 0.6125 MRR, 0.5496 Hit@1, and 0.7114 Hit@10, the best results among the re-rankers we evaluate; paired ensemble tests confirm the MRR and Hit@1 gains over SBERT, CTI-BERT, SecureBERT, CySecBERT, and a MiniLM cross-encoder. Diagnostics locate the remaining error in candidate generation, which caps Hit@10 at 0.7121.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.