mayaaah/cis5190-news-headlines

CIS 5190 News Headlines 是一个私有数据集，包含清洗过的新闻标题，用于新闻来源的二分类项目。每行数据包括标准化的标题文本、来源标签（NBC 或 FoxNews）、整数类别标签（0 对应 NBC，1 对应 FoxNews）、来源 URL（如果可用）、标准化为 YYYY-MM-DD 格式的发布日期（如果可用）以及来源文件出处。数据集规模为 1,499,660 行完整数据，并提供了 454,792 行的类别平衡子集。数据按时间划分为训练集（1,199,646 行，日期范围 2006-01-01 至 2022-07-20）、验证集（149,956 行，日期范围 2022-07-20 至 2024-03-01）和测试集（149,956 行，日期范围 2024-03-01 至 2026-04-28）。数据集旨在用于教育和研究实验，如标题分类、时间泛化、领域转移和数据集偏差分析，但存在局限性，例如仅包含标题和 URL、来源分布不平衡以及收集覆盖随时间变化。

CIS 5190 News Headlines is a private dataset containing cleaned news headlines collected for a binary news-source classification project. Each row includes a normalized headline text, source label (either NBC or FoxNews), integer class label (where 0 = NBC and 1 = FoxNews), source URL when available, publication date normalized to YYYY-MM-DD when available, and source file provenance. The dataset consists of 1,499,660 full rows and provides a class-balanced subset of 454,792 rows. It is temporally split into training set (1,199,646 rows, date range 2006-01-01 to 2022-07-20), validation set (149,956 rows, date range 2022-07-20 to 2024-03-01), and test set (149,956 rows, date range 2024-03-01 to 2026-04-28). The dataset is intended for educational and research experiments in headline classification, temporal generalization, domain shift, and dataset bias, with limitations such as containing only headlines and URLs, imbalanced source distribution, and varying collection coverage over time.