Predicting Users’ Personality on Social Media: A Comparative Study of Different Machine Learning Techniques

Abstract

The use of social media sites (SMSs) becomes ubiquitous worldwide as the number of users is noticeably increasing. This has led to exploiting such sites by market, business, and educational companies to deliver content that meets users' personal needs. However, this requires identifying users' personalities to respond to their individual preferences. This research aims at (1) analyzing users' posts on SMSs to predict their personality based on the Meyers-Briggs Type Indicator (MBTI) model, (2) comparing the performance accuracy of different preprocessing and data mining techniques, and (3) improving the prediction accuracy of users' personality types. The used dataset includes 8668 records in which each raw contains fifty posts. Three data mining techniques are applied namely, support vector machine (SVM), logistic regression (LR), and lightGBM. The findings suggest that lightGBM with the application of stemming, lemmatization, and grid search optimization as well as removing stop-words outperformed other techniques. The prediction accuracies for the four personality dimensions namely, Introversion-Extroversion (I-E), Intuition-Sensing (NeS), Feeling-Thinking (F-T), and Judging-Perceiving (J-P) are 100.0%. The research outcomes are promising as the four dimensions of MBTI have been identified effectively. Such outcomes are also compared with earlier research on personality prediction. This study can help SMSs providers, businesses, and educational institutions adapt their online sites based on users’ posts, tweets, and comments that can be used to predict their personality behavior.