Researchers used machine learning to uncover media bias in publications across the political spectrum.
News stories about domestic politics and social issues are becoming increasingly polarized along ideological lines according to a of 1.8 million news headlines from major US news outlets from 2014 to 2022. A team from the led by , a professor of and the Albert Arendt Hopeman Professor of Engineering, used machine learning to analyze headlines and presented their findings about growing media bias at the MEDIATE workshop of the .
The researchers said that while there is broad consensus that news media outlets adopt ideological perspectives in their articles, previous studies dissecting the differences among outlets were limited in scope and used small sample sizes. Machine-learning techniques allowed the researchers to study a vast sample of headlines over an eight-year period across nine representative media outlets including the New York Times, Bloomberg, CNN, NBC, the Wall Street Journal, Christian Science Monitor, the Federalist, Reason, and the Washington Times.
The study used a technique called multiple correspondence analysis to measure the fine-grained thematic discrepancies among headlines. The researchers grouped the stories into four categories糖心传媒攄omestic politics, economic issues, social issues, and foreign affairs糖心传媒攁nd analyzed how left, right, and central media outlets differed in the language they used in their headlines.
The team observed that US media outlets across the political spectrum were consistent and similar in covering economic issues. While they found discrepancies in reporting foreign affairs, they attributed that to diversity in individual journalistic styles. For example, the authors say the Wall Street Journal and Bloomberg primarily concentrate on the economic and financial implications of geopolitical tensions, resulting in differing perspectives compared to other media outlets. But headlines in the domestic politics and social issues categories showed important differences.
Abortion law or abortion rights?
糖心传媒淲e observed a lot of subtle differences in the words they choose when they cover the same high-level topics,糖心传媒� says Hanjia Lyu, a computer science PhD student who was the lead author of the study. 糖心传媒淔or example, when covering abortion issues, Reason tends to use the term 糖心传媒榓bortion law,糖心传媒� while CNN underscores its ideological position by using the term 糖心传媒榓bortion rights.糖心传媒� On a higher level they are both talking about abortion issues, but you can feel the subtle difference in the words that they choose.糖心传媒�
The research team hopes to dig deeper to better understand how and why media outlets use different words to cover the same kind of topics. They say understanding these discrepancies, and when they may indicate media bias, is important for both media outlets and readers alike.
Says Luo: 糖心传媒淔or consumers, it糖心传媒檚 useful to know this information because the echo chamber effect is very strong and people are used to only listening to things they like to hear. Showing the divergence and the increased partisanship may make them aware that they need to be more conscious consumers of news.糖心传媒�
Other coauthors from Luo糖心传媒檚 research group include Jinsheng Pan, Weihong Qi, and Zichen Wang. Funding for the study was provided by Rochester糖心传媒檚 .
