Prediction of image popularity over time on social media networks

This paper introduces a new problem that investigates image popularity on social networks. It is referred to as the stability of image popularity. It focuses on whether the popularity of a certain image on social networks will change over time or not. In order to investigate what drives the changes of popularity, we employ three features of social media data, i.e., social context, image semantics, and an image's early popularity. To predict the changes of images popularity, we collected 69,000 images from Instagram and kept track of their popularity for one month. For the popularity measure, we use Pareto principle. Using a GaussianNaive Bayes classifier, we predict the changes of image popularity, and the results show that the semantic feature is the only one that predicts the changes in the popularity of an image over time.