GARAGE dataset
收藏资源简介:
GARAGE数据集是由高丽大学网络安全学院主导构建的汽车网络安全知识库,旨在为基于大语言模型的攻击图生成提供结构化数据支持。该数据集整合了12,786个CVE漏洞描述和140份安全事件文档,数据来源于AutoSec-Timeline仓库及公开技术文档,经过多阶段处理形成标准化知识表示。创建过程采用混合检索增强生成框架,通过四大语言模型进行实体抽取、关系提取和事件结构化,最终构建符合STIX 2.1标准和Auto-ISAC ATM框架的知识图谱与向量存储。该数据集主要应用于汽车网络安全领域,支持自动化威胁分析与风险评估,能够帮助识别跨电子控制单元的链式漏洞,提升车辆级攻击路径的生成准确性与效率。
The GARAGE Dataset is an automotive cybersecurity knowledge base primarily developed by the School of Cybersecurity at Korea University, aiming to provide structured data support for attack graph generation powered by large language models (LLMs). This dataset integrates 12,786 CVE vulnerability descriptions and 140 cybersecurity incident documents sourced from the AutoSec-Timeline repository and publicly available technical materials, and has undergone multi-stage processing to form standardized knowledge representations. Its development adopts a hybrid retrieval-augmented generation framework, leveraging four large language models to perform entity extraction, relation extraction and event structuring. Finally, it constructs knowledge graphs and vector storage that comply with the STIX 2.1 standard and the Auto-ISAC ATM framework. Mainly applied in the automotive cybersecurity field, this dataset supports automated threat analysis and risk assessment, helping identify chained vulnerabilities across electronic control units (ECUs) and improving the accuracy and efficiency of vehicle-level attack path generation.
- 1GARAGE: Characterizing the Automation Boundary in LLM-based Attack Graph Generation高丽大学·网络安全学院; 三星电子; 国防发展局 · 2026年



