Social media platforms such as Facebook have been a crucial web traffic source for content providers. Content providers build websites and apps to publish their content and attract as many readers as possible. More readers mean more influence and revenue through advertisement. As Internet users spend more and more time on social media platforms, content websites also create social media presence, such as Facebook pages, to generate more traffic and thus revenue from advertisements. With so much content competing for limited real estate on social media users’ timelines, social media platforms begin to rank the contents by user engagements of previous posts. Posting content to social media that receives little user interaction will hurt the content providers’ future presence on social media. Content websites need to consider business sustainability when utilizing social media, to ensure that they can respond to short-term financial needs without compromising their ability to meet their future needs. The present study aims to achieve this goal by building a model to predict the advertisement revenue, which is highly correlated with user engagements, of an intended social media post. The study examined combinations of classification methods and data resampling techniques. A content provider can choose the combination that suits their needs by comparing the confusion matrices. For example, the XGBoost model with undersampled data can reduce the total post number by 87%, while still making sure that 49% of the high-performance posts will be posted. If the content provider wants to make sure more high-performance posts are posted, then they can choose the DNN(Deep Neural Network) model with undersampled data to post 66% of high-performance posts, while reducing the number of total posts by 69%. The study shows that predictive models could be helpful for content providers to balance their needs between short-term revenue income and long-term social media presence.
