iherman10/cis-5190-news-classifier-expanded-headlines

Expanded News Headlines (Fox vs NBC)

A dataset of recent U.S. news headlines from Fox News and NBC News, built for binary news-source classification (Track B of UPenn CIS 5190's final project).

Contents

• —~30k headlines spanning Nov 2025 – Apr 2026
• —Two columns:
• —url — original article URL
• —headline — the article headline as published

How it was built

Started from a course-supplied URL list scraped directly from each outlet, then expanded via topic- and date-scoped Google News RSS queries (site:foxnews.com / site:nbcnews.com, one query per (topic, day)) to broaden topical and temporal coverage. After scraping, headlines were:

• —exact-deduped on (headline, source) and again on a normalized key,
• —per-source downsampled so the combined Fox/NBC counts stay balanced,
• —merged with the original base set (skipping any headline already present).

Additional EDA-driven cleaning (Spanish-language drop, 25–140 character bounds, ≥4-word minimum, removal of pagination/footer artifacts like - Page N and - FOX News Radio, dedupe) is applied at training time by the project's preprocess.prepare_data().

Intended use

Educational use only — built as a course project to study textual style differences between two U.S. news outlets. Not curated for sentiment, factuality, or balanced topical coverage, and not intended for downstream applications beyond classroom modeling.