pengyueli/RAF_FolkTables

RAF-Ready Folktables是一个用于公平聚类研究的数据集，专门为评估RAF（检索增强数据集组装框架）而处理。该框架旨在通过数据层面的解决方案改善公平聚类：在聚类之前，从外部数据源检索和选择真实样本，以减少少数群体在语义区域中的代表性不足。数据集基于Folktables构建，后者源自美国社区调查公共使用微数据样本（ACS PUMS），并经过处理以支持表格数据领域的检索增强公平聚类设置。处理后的文件包括初始查询数据集（D_q）和外部源文件，用于作为候选数据源，还包含源元数据、嵌入和敏感群体标签，以运行RAF实验。该数据集专注于解决原始数据中少数群体分布不均的问题，通过增强数据来改善多数与少数群体在嵌入空间中的分布对齐。RAF-Ready Folktables is a dataset for fair clustering research, specifically curated to evaluate the RAF (Retrieval-Augmented Dataset Assembly Framework). This framework aims to improve fair clustering through data-level solutions: before clustering, retrieve and select real samples from external data sources to mitigate the underrepresentation of minority groups in semantic regions. The dataset is built upon Folktables, which is derived from the American Community Survey Public Use Microdata Samples (ACS PUMS), and has been processed to support retrieval-augmented fair clustering scenarios in the tabular data domain. The processed files include the initial query dataset (D_q) and external source files acting as candidate data sources, as well as source metadata, embeddings, and sensitive group labels for conducting RAF experiments. This dataset targets the uneven distribution issue of minority groups in the original data, and improves the distribution alignment between majority and minority groups in the embedding space via data augmentation.