Prediction of Average and Perceived Polarity in Online Journalism

We predicted the average and perceived journalistic objectivity in online news articles based on the text of the article and data collected on the individual user reading the article, in order to better understand both how journalistic bias relates to the text of an article alone, and how reader demographics and political leanings influence the perceived objectivity of a given article. Using survey data on the perceived polarity of 284 articles ranked by multiple readers using the Mechanical Turk crowdsourcing platform, we applied supervised machine learning algorithms including L2-regularized L2loss SVM, Naive Bayes, and Linear Regression to predict the mean polarity for each article and the perceived polarity of that article by any given reader. We were able to identify the words most strongly correlated with journalistic bias, predict a binary classification of mean and perceived bias for each article with relatively high accuracy, and determine the degree to which the article text alone and reader demographic information impacts perception of journalistic bias.