AttackSeqBench
收藏资源简介:
AttackSeqBench是一个针对评估大型语言模型在理解网络安全威胁情报报告中攻击序列方面的基准数据集。该数据集由新加坡国立大学构建,包含了408个经过精心构建的攻击序列,这些序列基于真实世界的CTI报告。数据集通过自动化数据生成管道构建,包括攻击序列构建、问题生成和自我完善三个阶段,旨在系统评估LLM在分析攻击序列方面的能力,并推动其在现实世界网络安全操作中的应用。
AttackSeqBench is a benchmark dataset dedicated to evaluating large language models' (LLMs') capability in comprehending attack sequences from cybersecurity threat intelligence (CTI) reports. Constructed by the National University of Singapore, this dataset comprises 408 meticulously curated attack sequences sourced from real-world CTI reports. It is developed through an automated data generation pipeline encompassing three phases: attack sequence construction, question generation, and self-refinement. Its core objective is to systematically assess LLMs' ability to analyze attack sequences and advance their practical applications in real-world cybersecurity operations.




