NewsSpectrum

Overview

NewsSpectrum is a news corpus balanced across political media-bias categories. It was built for DiversiNews, the news-retrieval demo of diversity-aware inner-product search covered in Diversity-Aware k-Maximum Inner Product Search Revisited. Every article carries its outlet’s bias rating, so a retrieved set can be scored on how far it spreads across the political spectrum, as well as on relevance.

Construction

PropertyValue
SourceURLs from Reddit submissions in the Pushshift dump, up to July 2022, with at least 10 upvotes
Articles250,000
News sources961
Bias categoriesLeft, Lean Left, Center, Lean Right, Right — 50,000 articles each
Label sourceAllSides media-bias ratings, matched by outlet domain, scored −2 to +2

What the labels are — and are not

The bias labels describe outlets, not individual articles. The dataset’s own documentation says they are not independently annotated article-level ideology or factual-accuracy labels. A diversity score computed on them, such as the mean pairwise bias difference used in DiversiNews, measures a spread of sources, which is only a proxy for a spread of viewpoints. The corpus is also concentrated on US politics. The DiversiNews authors note this but argue the method does not depend on it, because the labels are used only for evaluation and never for encoding or retrieval.

Use

  • DiversiNews — the evaluation corpus, encoded with Sentence-BERT, AnglE and LLaMA 2 embeddings and searched with DkMIPS at k = 10.
  • DkMIPS journal version — the accompanying repository lists NewsSpectrum (249,000 vectors, encoded with AnglE UAE-Large-V1) as one of its two text-retrieval evaluation sets.

Availability

Published in the DiversiNews repository via Git LFS. The MIT license covers the code and the curated metadata and labels; the article texts remain the property of their publishers and are provided for research use.

  • LocalNews — another diverse-news benchmark from the same research group, aimed at aspect coverage rather than political spread