Fraud-R1
收藏资源简介:
Fraud-R1是一个新颖的双语(英语和中文)基准,旨在评估LLM在五种现实世界场景(欺诈服务、冒名顶替、网络钓鱼诈骗、虚假招聘和在线关系)中防御欺诈和钓鱼诱导的能力。该数据集包含从社交媒体、新闻报告、讲座材料和先前单一任务欺诈数据集中手动收集的全面欺诈案例。数据集分为两个子集:FP-base和FP-levelup,后者是基于前者的规则增强数据集,用于多轮对话设置。
Fraud-R1 is a novel bilingual (English and Chinese) benchmark designed to evaluate the ability of Large Language Models (LLMs) to defend against fraud and phishing inducements across five real-world scenarios: fraudulent services, identity impersonation, phishing scams, fake recruitment, and online relationships. This dataset contains comprehensive fraud cases manually collected from social media, news reports, lecture materials, and prior single-task fraud datasets. It is divided into two subsets: FP-base and FP-levelup. The latter is a rule-enhanced dataset based on the former, tailored for multi-turn conversation settings.




