Solutions of Brand Posts on Facebook to Increase Customer Engagement Using the Random Forest Prediction Model

Abstract

This paper aims to predict customer engagement behaviour (CEB), i.e. likes, shares, comments, and emoji reactions, on company posts on Facebook. A sample of 1109 brand posts from Facebook pages in Lithuania was used. The Random Forest method was used to train models to predict customer engagement behaviour based on features including time frame, content, and media types of brand posts. The data was used for training nine binary classification models using the Random Forest method, which can predict the popularity of a company’s posts. In terms of social score, accuracy of likes, comments, and shares varied from 68.4% (likes on a post) to 84.0% (comments on a post). For emotional responses, accuracy varied from 65.6% (‘wow’ on a post) to 82.5% (‘ha ha’ on a post). The data was collected from one single media platform and country, and encompassed emotional expressions at an early stage on Facebook. The findings of Random Forest prediction models can help organisations to make more efficient solutions for brand posts on Facebook to increase customer engagement. This paper outlines the first steps in creating a predictive engagement score towards diverse types of brand posts on Facebook. The same approach to features of brand posts might be applied to other social media platforms such as Instagram and LinkedIn.