Using Linguistic Features to Predict Social Media Engagement: Proposing an Approach Based on Machine Learning and Natural Language Processing

Social media customer engagement plays a crucial role in B2B companies’ growth and success. Thus, knowing if a social media post will engage customers can be of great value for B2B companies. In this research, using 51,615 Facebook posts from 121 B2B companies, we develop machine learning-based models that classify social media posts into successful and unsuccessful posts in engaging the customers. First, using a score obtained from the number of customer reactions to a post (likes, comments, and shares), we label the data and extract features from the most frequent words of the training dataset. Then we employ three well-known supervised machine learning algorithms (support vector machine, Naïve Bayes, and multi-layer perceptron) to classify the labeled data. The best models show moderate performance in predicting the success of a Facebook post, achieving an accuracy and F1-score of 72.2%. Additionally, our results show that the presence of inspirational and positive words and the words related to time and date has a positive impact on user engagement.