2026-10-06T09:28:54Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33949330 2026-09-21T07:11:40Z category_28921 item_type_3 month_year_09_2026

Reproducibility supplement for "Evaluating ColBERT for Cyber Threat Intelligence: Supervision and Coverage in ATT&CK Mapping" Eshan Ahmad (25080052) System and network security ColBERT cyber threat intelligence MITRE ATT&CK information retrieval late interaction cybersecurity <p dir="ltr">Reproducibility supplement accompanying "Evaluating ColBERT for Cyber Threat Intelligence: Supervision and Coverage in ATT&CK Mapping" by Eshan Ahmad, Department of Computer Science, University of Nevada, Las Vegas. This deposit contains the source-separated ATT&CK technique-mapping benchmark, a designed 50-label exclusion evaluation, experiment code, saved model scores, statistical checks, trained model states, and the numerical supplement. The study evaluates ColBERT late interaction alongside lexical, learned-sparse, dense, and supervised classification baselines. It evaluates technique retrieval as a proxy for the retrieval stage of a security RAG system; generated answers are not evaluated. Start with README.md. The experiment archive contains code, data, results and verification instructions. The trained-model archive is optional for replaying saved-score statistics. The LaTeX archive contains the manuscript and supplement sources. The analysis includes 74 primary/held-label metric sets and one 65-contrast Holm family. A separate exploratory BGE diagnostic compares training-order shuffling and title-negative filtering on 542 queries for 50 excluded labels. The supplement documents the checks and their scope; saved-score verification is distinct from independently repeating all model training. Licensing scope: CC BY 4.0 applies to original text, documentation and results. MIT applies to original experiment and analysis code. Third-party material retains its existing terms; these are component-specific licenses, not alternative licenses for every file. See LICENSES.md. This deposit supports a research manuscript; it does not assert IEEE publication or conference acceptance.</p> 2026-09-21T07:11:40Z Dataset Dataset 10.6084/m9.figshare.33949330.v1 https://figshare.com/articles/dataset/Reproducibility_supplement_for_Evaluating_ColBERT_for_Cyber_Threat_Intelligence_Supervision_and_Coverage_in_ATT_CK_Mapping_/33949330 CC BY 4.0