Abstract
Rappler, one of the leading online news publishers in the Philippines, obtains audience sentiment through the Mood Meter widget embedded in its articles. The Mood Meter allows a reader to select an article’s mood from a set of predefined moods (happy, inspired, amused, sad, angry, afraid, annoyed, don’t care) with varying polarities (positive mood or negative mood). Using machine learning algorithms, we created two classification models to predict a Rappler article’s dominant mood and polarity.
A total of 5,735 Rappler articles with metadata such as category, author, and mood ratings were used as the training set. The resulting models can predict the article polarity with 72% accuracy and the article’s dominant mood with 51% accuracy. These models can aid Rappler and other news publishers in content personalization, article engineering, and dynamic ad placement and pricing.
