Data-driven approaches into political orientation and news outlet discrimination: The case of news articles in South Korea

Abstract

With the advancement of the internet, the public now has easy access to news from various media outlets. However, a number of news outlets tend to report their content based on their political orientations or affiliations, which may compromise the objectivity of the news. This research used machine learning to analyze whether it is possible to tell the political orientation and news outlet apart based on news articles in South Korea. We collected a lot of news articles spanning over five years and used the text data. We chose major conservative and progressive news outlets and tried classifying them into two groups. We even looked into classifying articles by each news outlet. We used different machine learning methods like Logistic Regression, Random Forest Classifier, and eXtreme Gradient Boosting, and tried to improve the performance by combining these models. The research found that the combined model had high accuracy, up to 91.9% for binary classification of news outlet political orientations and up to 84.0% for classifying news outlets in multiple categories. This shows that you can determine the political leaning of news outlets based on their articles, highlighting the importance of considering bias in news outlets when evaluating information instead of solely relying on the article content.