Dataset for NER\/RE model training in vulnerability reports

This dataset is curated for Named Entity Recognition (NER), NER Categorization, and Relation Extraction (RE) tasks, focusing on cybersecurity vulnerability descriptions. It includes annotated text from vulnerability reports, labeling entities such as product names (PN), versions (V), and modifiers (MOD), and categorizing product names into applications (APP), hardware (HW), or operating systems (OS). Relations between entities are marked using special tokens [PN] and [V+MOD] to indicate associations with vulnerabilities. We sampled 5,000 vulnerability descriptions (3,000 pre-2019 and 2,000 post-2019) for balanced temporal representation. The dataset enables the development and evaluation of machine learning models for automated vulnerability analysis, facilitating precise identification and classification of security risks in software and hardware systems.