shaswatamitra/falcon-yara-cti-rule
收藏资源简介:
FALCON YARA CTI ↔ 真实规则数据集是一个网络安全基准数据集,包含网络威胁情报(CTI)描述及其对应的真实YARA规则,以及由大语言模型生成的诱饵规则(硬负样本相似规则)。该数据集是FALCON保留的测试集,用于评估大语言模型的规则生成能力,以及测试对比性微调句子编码器的检索性能。数据集包含916个CTI描述和5,106个诱饵规则,每个CTI平均有5.57个诱饵规则,范围在0到12之间。数据来源是从FALCON yara-rules微调语料库中确定性保留的10%测试分割,真实规则通过反转映射恢复。
FALCON YARA CTI ↔ Ground Truth Rule Dataset is a cybersecurity benchmark dataset that contains cyber threat intelligence (CTI) descriptions, their corresponding ground truth YARA rules, as well as decoy rules (hard negative similar rules) generated by Large Language Models (LLMs). This dataset is a test set retained by FALCON, which is used to evaluate the rule generation capability of LLMs and test the retrieval performance of contrastively fine-tuned sentence encoders. The dataset includes 916 CTI descriptions and 5,106 decoy rules, with an average of 5.57 decoy rules per CTI, ranging from 0 to 12. The data originates from the 10% test split deterministically retained from the FALCON yara-rules fine-tuning corpus, and the ground truth rules are recovered through inverse mapping.




