shadow-bench/SB-unlearning

SB-Unlearning是一个专门的数据集，旨在使用ShadowBench框架评估机器遗忘算法。它专注于高密度实体（如Elon Musk）的定向移除，并测量显式词汇遗忘与潜在概念关联之间的差距。数据集包含训练和评估语料库，用于机器遗忘目标的显式问答对和检测残留知识的潜在Shadow探针。数据集分为多个子集，包括elon_musk_forget、elon_musk_retain、elon_musk_shadow_forget和elon_musk_shadow_retain，每个子集有不同的分割和用途。数据集适用于问答任务，语言为英语，许可证为CC BY-SA 4.0。

SB-Unlearning is a specialized dataset designed to evaluate Machine Unlearning algorithms using the ShadowBench framework. It focuses on the targeted removal of high-density entities (e.g., Elon Musk) and measures the gap between explicit lexical forgetting and latent conceptual association. The dataset provides the training and evaluation corpora used for the unlearning objective, including explicit QA pairs and latent Shadow probes to detect residual knowledge. It is divided into several subsets, including elon_musk_forget, elon_musk_retain, elon_musk_shadow_forget, and elon_musk_shadow_retain, each with different splits and purposes. The dataset is suitable for question-answering tasks, is in English, and is licensed under CC BY-SA 4.0.