Title: Media bias detection using sentometrics
Authors: Jeroen Van Pelt - Vub (Belgium) [presenting]
Andres Algaba - Vrije Universiteit Brussel (Belgium)
Samuel Borms - Universite de Neuchatel (Switzerland)
Kris Boudt - Vrije Universiteit Brussel and VU Amsterdam (Belgium)
Abstract: Media bias can strongly influence the public opinion on important topics which in turn can ultimately affect peoples decision-making process. While the social sciences recognize the important impact of media bias, automated and scalable methods to detect textual bias in news articles are lacking. Media bias analysis tools are useful to quantify and visualize the relative difference between the coverage and reporting of a news fact by one news medium compared to its peers. We identify four dimensions of media bias through reporting, and define appropriate metrics for each dimension. These bias metrics can be constructed using a combination of sentiment analysis and econometrics, which we refer to as sentometrics. The reporting bias can then be computed by comparing these metrics to its peer articles that have been identified via a cosine similarity based matching tool. Finally, we empirically show the effectiveness of automated bias detection in media news articles.