Abstract: We study Facebook posts published by major news organizations in the 10-month period leading to the 2016 presidential election. Our goal is to explore the topics related to the two major party candidates, Hillary Clinton and Donald Trump, and identify the ones that engaged the Facebook users the most. The engagement is measured by the total number of reactions, comments, and shares. Using topic modeling with Linear Dirichlet Allocation (LDA) on the Facebook posts, we identify the top 10 topics related to each candidate and then assess the audience engagement for these topics across 10 different news organizations. We use Hierarchical Bayesian Models (HBMs) to analyze the data, which allow us to partially pool the information across different sources.
Data taken from: https://data.world/martinchek/2012-2016-facebook-posts
The extended version of our paper is available here: https://github.com/milkha/FBElec16/blob/master/FullPaper.pdf