Analyzing the Political Leaning of Electronic Media with regards to its Coverage of South Asia: A Machine Learning Approach

Abstract

There has been a lot of discussion about the political leanings of news. This paper proposes and implements an automated method to analyze the political leaning of electronic media with regards to its coverage of two countries in South Asia: India and Pakistan. We use an explicitly political text, the speeches of Congressmen for two time periods (1997 - 1999 and 2015 - 2017), to evaluate the political leaning of news articles and transcripts during this time. In addition, we also evaluate the change in political leaning over time from 2000 - 2015 and compare our results to previous empirical studies on the subject.

We trained multiple classifiers on our training data and found that Support Vector Machines performed best. We concluded that CNN and Washington were more likely to be liberal while Fox News and Wall Street Journal were likely to be conservative. Our results on political biases agreed with previous empirical studies on the matter. We also found out that the news coverage of the two countries, India and Pakistan, has been overall pretty similar in its political leaning. However, the conservativeness of the news has seen a steady increase over the last three years.

Our work establishes the validity of using SVM classifiers for mining the political leanings of text. It also confirms popular notions of the Democratic leanings of CNN and the Republican leanings of Fox News. It informs us of the political biases of electronic media towards South Asia and the impact of major events on the coverage of Pakistan and India. We believe that our work is a step forward in utilizing automated methods for understanding political biases.