Ideology Detection of Personalized Political News Coverage: A New Dataset

Abstract

Words selection, writing style, stories cherry-picking, and many other factors play a role in framing news articles to fit the targeted audience or to align with the authors' beliefs. Hence, reporting facts alone is not evidence of bias-free journalism. Since the 2016 United States presidential elections, researchers focused on the media influence on the results of the elections. The news media attention has deviated from political parties to candidates. The news media shapes public perception of political candidates through news personalization. Despite its criticality, we are not aware of any studies which have examined news personalization from the machine learning or deep neural network perspective. In addition, some candidates accuse the media of favoritism which jeopardizes their chances of winning elections. Multiple methods were introduced to place news sources on one side of the political spectrum or the other, yet the mainstream media claims to be unbiased. Therefore, to avoid inaccurate assumptions, only news sources that have stated clearly their political affiliation are included in this research. In this paper, we constructed two datasets out of news articles written about the last two U.S. presidents with respect to news websites' political affiliation. Multiple intelligent models were developed to automatically predict the political affiliation of the personalized unseen article. The main objective of these models is to detect the political ideology of personalized news articles. Although the newly constructed datasets are highly imbalanced, the performance of the intelligent models is reasonably good. The results of the intelligent models are reported with a comparative analysis.