Prediction of Thinking-Feeling personalities of movie characters by text data using Machine Learning algorithms

Abstract

This paper is going to explore the difference between the vocabularies used by Thinking and Feeling personalities. To find out this we used the Machine Learning algorithm Naïve Bayes which showed the best accuracy in comparison with others. The concept was motivated by essays of scholars when they submitted the first time at university and to get the full psychological portrait of the student only by given text. To train the model we used a labeled dataset that was collected through a forum with real persons. This dataset contains the type of the person and their posts in social media. To test the model using another dataset which contains information about movie characters and their speech used in the movie. Psycho-type was described by Myers-Briggs Type Indicators (MBTI) which is one of the most popular typologies. To achieve better accuracy of prediction we trained the model separately for Thinking and Feeling predictors. Overall, we achieved better accuracy than previous studies and showed the difference between the vocabularies used by Thinkers and Feelers.