NewsCorpus-LLMExplain

We introduce NewsCorpus covering contemporary events curated from AllSides from July 21, 2012 – July 23, 2024.The dataset contains 17,166 articles grouped into 5,722 news round-ups across 64 topical categories.

Data were collected using a public scraping tool. The scraping script was adapted due to updates in the AllSides HTML structure.

Each row/ roundup contains three parallel news stories representing different political perspectives:

• Left or Left-leaning media
• Center media
• Right-leaning or Right media

Citation

If you use this dataset or build upon it in your research, please cite the following paper:

Wang, Q., Khatiwada, P., Chouhan, A., Mahesh, A., Mwaria, J., Tran, D. D., ... & Mauriello, M. L. (2026). " The explanation makes sense": An Empirical Study on LLM Performance in News Classification and its Influence on Judgment in Human-AI Collaborative Annotation. arXiv preprint arXiv:2602.19690.

@article{wang2026explanation,
  title={"The explanation makes sense": An Empirical Study on LLM Performance in News Classification and its Influence on Judgment in Human-AI Collaborative Annotation},
  author={Wang, Qile and Khatiwada, Prerana and Chouhan, Avinash and Mahesh, Ashrey and Mwaria, Joy and Tran, Duy Duc and Barner, Kenneth E and Mauriello, Matthew Louis},
  journal={arXiv preprint arXiv:2602.19690},
  year={2026}
}

📄 NewsCorpus17K.csv

👉 Checkout preview: NewsCorpus17K_sample100.csv

Core Metadata

• Date — publication date of the roundup
• Topic — categorical topic label
• Title of Headline Roundup — AllSides roundup headline
• url_story — AllSides roundup link

Left/Left-Leaning Story

• left_story_title — article headline
• left_story_url — article URL
• left_story_source — publisher name
• left_story_leaning — political bias label
• left_story_text — preview or snippet text

Center Story

• center_story_title — article headline
• center_story_url — article URL
• center_story_source — publisher name
• center_story_leaning — political bias label
• center_story_text — preview or snippet text

Right-Leaning/Right Story

• right_story_title — article headline
• right_story_url — article URL
• right_story_source — publisher name
• right_story_leaning — political bias label
• right_story_text — preview or snippet text

Major topic groups include:

• Elections and Democracy: 2024 Presidential Election, Elections, Voting Rights and Voter Fraud, Campaign Finance, Polarization
• US Politics and Law: Politics, Supreme Court, Civil Rights, Federal State and Tribal Powers, Free Speech, Privacy
• Justice and Safety: Criminal Justice, Violence in America, Gun Control and Gun Rights, Terrorism, Sexual Misconduct
• Health and COVID: Healthcare, Public Health, Coronavirus, Life During COVID-19, COVID-19 Misinformation, Safety and Sanity During COVID-19
• Economy and Business: Economy and Jobs, Business, Banking and Finance, Taxes, Trade, Housing and Homelessness
• International Affairs: World, Foreign Policy, The Americas, Middle East, China, Russia, Ukraine War, Defense and Security
• Society and Rights: Race and Racism, LGBTQ Issues, Abortion, Education, Family and Marriage, Religion and Faith, Inequality
• Science and Technology: Science, Technology
• Environment and Energy: Environment, Climate Change, Sustainability, Energy, Water and Oceans
• Media and Information: Media Industry, Media Bias, Fake News, Facts and Fact Checking, Common Ground
• Culture and Sports: Culture, Arts and Entertainment, Sports
• People and Figures: Joe Biden, Donald Trump

Political bias is assigned at the publisher level using AllSides ratings. Each article inherits the political leaning of its publisher. Prior work shows approximately 97% agreement between publisher-level and article-level bias.

---

LLMExplain

We selected 60 recent news articles across 20 headline round-ups from NewsCorpus17k.csv. Each record includes aligned left, center, and right news perspectives enriched with GPT-generated political ideology predictions and two types of explanations (brief and detailed). The explanations are intended to provide transparent reasoning behind the model’s predicted labels. This subset is designed for use in human annotation tasks and evaluation studies.

📄 Brief Condition: LLMExplanationSubset-Brief.csv

📄 Detailed Condition: LLMExplanationSubset-Detailed.csv

Additional Features

• left_story_GPT_pred — GPT-predicted political leaning
• left_story_GPT_explanation — GPT-generated explanation of prediction
• center_story_GPT_pred -— GPT-predicted political leaning
• center_story_GPT_explanation — GPT-generated explanation of prediction
• right_story_GPT_pred — GPT-predicted political leaning
• right_story_GPT_explanation — GPT-generated explanation of prediction

---

Notes

This dataset contains publicly available news metadata collected for research purposes.
Users should respect original publisher copyright and terms of service.

For questions, contact: kylewang@udel.edu

License

This project is licensed under the MIT License. See LICENSE.md for details.