Hi everyone,<\/p>\n
I\u2019m sharing a metadata-only dataset of 7,000 news articles (extracted from a larger 700k core) designed specifically for NLP feature engineering and Media Intelligence. Instead of just standard sentiment (Positive\/Negative), I\u2019ve focused on “Narrative Alpha”, structural signals that quantify how a story is being told.<\/p>\n
Why this is useful: If you’re building news classifiers, bias detectors, or financial sentiment models, standard text often isn’t enough. This set provides deterministic linguistic metrics you can’t get from a standard scrape.<\/p>\n
What\u2019s Inside (22 Columns):<\/p>\n
- \n
- Structural Metrics: Passive Voice Ratio, Sentence\/Word Counts.<\/li>\n
- Narrative Signals: Hedging Rate (uncertainty cues), Claim Density per 1k words.<\/li>\n
- Credibility & Alignment: Headline-Body Alignment Score, Primary Source Ratio (attribution).<\/li>\n
- Traditional Labels: Topic, Political Orientation, Bias Strength, Credibility Level.<\/li>\n<\/ul>\n
Technical Specs:<\/p>\n
- \n
- Format: Tabular CSV (Clean, no text blobs to protect legal\/copyright).<\/li>\n
- Usability: 10.0\/10.0 on Kaggle (fully documented columns).<\/li>\n
- License: CC BY 4.0 (Open for research\/commercial use).<\/li>\n<\/ul>\n
Link: Kaggle <\/a><\/p>\n
AMA about the methodology or the pipeline!<\/p>\n<\/div>\n
submitted by \/u\/Queasy_System9168 <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
