Popularity Prediction of Images and Videos on Instagram

We live in a world surrounded by numerous social media platforms, applications and websites which produce various texts, images and videos (posts) daily. People share their moments with their friends and families via these tools to keep in touch. This extensiveness of social media has led to an expansion of information in various forms. It is difficult to imagine someone totally unfamiliar with these concepts and not having posted any content on a platform. All users, ranging from individuals to large companies, want to get the most of their audiences' attention. Nevertheless, the problem is that not all these posts are admired and noticed by their audience. Therefore, it would be important to know what characteristics a post should have to become the most popular. Studying this enormous data will develop a knowledge from which we can understand the best way to publish our posts. To this end, we gathered images and videos from Instagram accounts and we used some image/video context features to predict the number of likes a post obtains as a meaning of popularity through some regression and classification methods. By the experiments with 10-fold cross-validation, we get the results of Popularity Score prediction with 0.002 in RMSE and Popularity Class prediction with 90.77% accuracy. As we know, this study is the first exploring of Iranian Instagram users for popularity prediction.