AMPEL: An Approach for Machine-learning Based Prediction and Evaluation of the Learned Success of Social Media Posts

In this paper, we present AMPEL, a system that assists social media managers in creating successful posts for company pages on social media platforms. The AMPEL workflow classifies posts as either successful or unsuccessful, using Facebook as an example. The system makes a prediction of success for a new post during its creation, prior to its publication, which is a major advantage in comparison to existing systems. Posts that are classified as unsuccessful can be revised by the author until the prediction is successful. The system also evaluates previously published posts for success, allowing for comparison between the predicted success and actual success achieved. The two classification models are built using Random Forest, XGBoost, and neural network-based classification algorithms. The system is evaluated using two separate corpora of posts from different industries. We also demonstrate a prototype of AMPEL and show that it achieves good results with very reliable predictions of success. Our evaluation shows that AMPEL can replace manual review by a human social media manager in many applications.