Mohamed Eltoukhi
CVE Intelligence Extraction: Fine-Tuning Local Large Language Models for Structured Vulnerability Data.
Rel. Fulvio Valenza, Daniele Bringhenti, Riccardo Sisto. Politecnico di Torino, Corso di laurea magistrale in Cybersecurity, 2026
|
Preview |
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (3MB) | Preview |
Abstract
The number of published Common Vulnerabilities and Exposures (CVE) records continues to grow, but much of their useful information is still written in free-form text. People can read these descriptions, whereas automated security tools work better with clearly defined fields. Hand-written parsing rules offer one solution, but they can fail when descriptions differ in wording or level of detail. An open-weight language model provides an alternative for processing these descriptions. The proposed extractor converts raw CVE records into structured JavaScript Object Notation (JSON) directly on the user's machine. Its target format, VIPER (Vulnerability Intelligence Parsed Extraction Record), organizes affected products, weaknesses, conditions required for exploitation, and possible consequences.
A larger teacher model generated 989 CVE-to-VIPER examples, avoiding the need to label hundreds of records by hand
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Corso di laurea
Classe di laurea
URI
![]() |
Modifica (riservato agli operatori) |
