Predicting Popularity of Facebook Videos Through Visual Features Using Support Vector Machine Classifier

With the popularization of social networks, the sharing and consumption of content in video format becomes easier. Understanding what makes a video popular and being able to predict its popularity in number of views is useful for both content creators and advertising. In this work, we explore visual features extracted from 1,820 Facebook videos in order to predict whether they will reach more than a certain number of views on the seven days after publication. For this purpose, we used Support Vector Machine with Gaussian Radial Basis Function classification model. Using only visual features as predictors, the model with Video Characteristics and Rigidity features combined reached Kappa of 0.7324, sensitivity of 0.8930, and positive predictive value of 0.8930.