Penn Computational Social Sciences Lab\u2019s Media Bias Detector Project\u2019s scientific lead spoke with The Daily Pennsylvanian about how the platform uses artificial intelligence models to classify politically relevant news coverage.<\/p>\n
The detector analyzes<\/a> articles from multiple news publishers daily and categorizes them based on their political affiliation. Samar Haider, a fifth-year graduate student and the project\u2019s scientific lead, told the DP that the detector tracks 21 publishers \u201cthat cover a broad spectrum of left to right leaning ideologies.\u201d<\/p>\n Haider added that the software is intended to be used by a broad audience and \u201cgives news readers an overview of what\u2019s being talked about in the news and how it\u2019s being talked about.\u201d It also identifies \u201clong term patterns, both within publishers and across publishers that are markers of bias.\u201d\u00a0<\/p>\n \u201cThe target is essentially anyone who is interested in or consumes news,\u201d Haider said. \u201cThis can range from the average person to a journalist or a researcher on media and communication.\u201d\u00a0<\/p>\n The project was funded by 1989 Wharton graduate Richard Mack and is the result of a collaboration between CSSLab Director Duncan Watts, Managing Director Jeanne Ruane, and a team of researchers.<\/p>\n The program \u2014\u00a0which was launched prior to the 2024 General Election \u2014 initially collected information from 10 publications. In January, the project was planning<\/a> to expand to cover 22 publications.\u00a0<\/p>\n To collect data, the detector<\/a> extracts information from each selected publication\u2019s website five times a day, identifying the 20 \u201cmost prominent\u201d articles and evaluating them on different metrics. The platform employs GPT models, machine learning, and human raters to classify the article, identify political preferences, and detect tone on a scale from \u201cVery Negative\u201d to \u201cVery Positive.\u201d\u00a0<\/p>\n Watts, the twenty-third Penn Integrates Knowledge Professor, wrote<\/a> that the detector was inspired by the \u201cdivisiveness pervading popular media\u201d that \u201cwe\u2019ve all experienced.\u201d\u00a0<\/p>\n \u201cOur hypothesis is that this is exacerbated not by lies or \u2018fake news\u2019 but by bias,\u201d he added.<\/p>\n Haider emphasized that the model\u2019s consistency over time is the \u201conly way that the trends that we extract over time will make sense,\u201d because every article is evaluated \u201con the same merits.\u201d<\/p>\n