iiinova — where your digital self keeps evolving

Samar Haider

@samarhaider.bsky.social

Computational social scientist studying the media information ecosystem. Now postdoc @ Northwestern, previously PhD @ Penn. samarh.github.io

Posts Replies Media

Samar Haider @samarhaider.bsky.social · 16/09/2026

This project has been a huge undertaking & would not have been possible without a great team of collaborators at @csspenn.bsky.social @penn-mediated.bsky.social: @atohidi.bsky.social, @jennyshwang.bsky.social, Timothy Dorr, @davmicrot.bsky.social, Chris Callison-Burch, and @duncanjwatts.bsky.social!

010

Samar Haider @samarhaider.bsky.social · 16/09/2026

We hope our work helps improve both academic understanding and public awareness of media bias dynamics through an empirical, data-driven approach. Read our paper: science.org/doi/10.1126/... Browse our dashboard: mediabiasdetector.seas.upenn.edu Use our data and code: github.com/Watts-Lab/me...

140

Samar Haider @samarhaider.bsky.social · 16/09/2026

We release a dataset of 140K+ annotated news articles from 10 major publishers to enable researchers to build on this framework. Our interactive web dashboard complements this dataset by letting users dynamically explore patterns in news coverage.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

Most news at the top of publishers’ homepages is political, and nearly all of it is negative. These are just a few examples of the research questions we can start to address with this framework.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

Political lean is not one-dimensional, with substantial variation across topics even within a publisher.

111

Samar Haider @samarhaider.bsky.social · 16/09/2026

We also find that news publishers talk more about the opposition party than the one they are more ideologically aligned with.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

By labeling headlines and articles independently, we observe a systematic misalignment in which headlines are coded as more right-leaning than the articles themselves, especially for economic news.

120

Samar Haider @samarhaider.bsky.social · 16/09/2026

Using this data, we study a range of patterns in news coverage and present illustrative findings that showcase what we can learn from this framework. We find that the media covered the election horse race more than 4x as much as all substantive policy topics combined in 2024.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

Our validation experiments show that LLM annotations align closely with human judgements across these coding tasks.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

We use LLMs to label each article at multiple levels of granularity, including topic, tone, political lean, sentence composition, quoted individuals, event clusters, and more.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

The Media Bias Detector is a scalable computational framework that regularly collects and annotates thousands of news articles from major publishers in near-real-time.

110

Samar Haider @samarhaider.bsky.social · 16/09/2026

New paper out in Science Advances @science.org! News organizations shape public understanding through the choices they make about what to cover and how to cover it. How can we measure these differences across publishers systematically and at scale? We present the Media Bias Detector!

1216

Samar Haider @samarhaider.bsky.social · 08/09/2026

I'm thrilled to share that I defended my PhD at @upenn.edu! I've had the time of my life working with my amazing advisors and all the wonderful people at @csspenn.bsky.social over the past several years. Incredibly grateful for the journey and everyone who made it so special!

021