To Share or Not to Share: Investigating Drivers for Sharing Online News Using Automated Machine Learning and Probabilistic Modeling

Abstract The article leverages advanced machine learning to investigate what drives sharing behavior for online news content. To this end, it investigates a dataset of 39,797 pieces of individual news and uses 59 different features to outline the key influences on the number of shares. Initially, an automated machine learning framework is applied to choose the optimal model among 1,000 contenders and then this model is used to study the likelihood of sharing further. Causality links are investigated in more detail by recourse to a Bayesian Belief Network, which elucidates the transmission mechanisms and the direct quantitative effect of relevant predictors. Overall results show that the volume and uniqueness of content, appropriate keywords, and the article’s position in the information network are all important predictors for sharing. Conversely, negative polarity is shown to be the most important blocker.