nyhetsdataset

Nyhetsdataset is a corpus of news articles assembled for research and development in natural language processing, text mining, and data-driven journalism. The collection is typically focused on Swedish-language news, with the possibility of including items from additional languages or translated content to support multilingual experimentation. The dataset is intended to enable tasks such as classification, summarization, information extraction, sentiment analysis, and trend detection over time.

The content is organized with articles accompanied by metadata. Common fields include article_id, title, lede or summary, full text, publication date, source or outlet, author, and category or topic. Additional metadata may cover language, geographic scope, and confidence scores for automated annotations. Data formats are typically JSON, JSONL, or CSV, sometimes with separate index files for fast lookup. The corpus may be released in versions or dumps that reflect updates from participating outlets and syndication services.

Collection and licensing practices vary. Articles are gathered from public-facing news websites, feeds, or licensed providers, and are subject to the terms of use of each source. Licensing is often open or research-friendly (for example, CC BY or CC0, or restricted by source agreements). Deduplication, content cleaning, and redaction of sensitive information are common preprocessing steps.

Annotations and enhancements are sometimes included, such as topic labels, editorial categories, or linguistic marks (tokenization, lemmatization). Some releases offer sentiment or bias indicators, or event tagging, produced either automatically or by manual curation.

Applications include benchmarking NLP models, building news-aware recommender systems, and studying media coverage and discourse. Users should consider source bias, coverage gaps, copyright restrictions, and privacy concerns when employing nyhetsdataset.