Transformer-based Approaches for Personality Detection using the MBTI Model

Abstract

Personality Detection is a well-known field in Artificial Intelligence. Similar to Sentiment Analysis, it classifies a text in various labels that denote common patterns according to personality models such as Big-5 or Myers-Briggs Type Indicator (MBTI). Personality detection could be useful for recommendation systems, improvements in health care and counseling, forensics, job screening, to name a few applications. Most of the works on personality detection use traditional machine learning approaches which rely on open dictionaries and tokenizers resulting in low performance and replication issues. In contrast, Deep Learning Transformer models have gained popularity for their high performance. In this research, we propose several Transformer approaches for detecting personality according to the MBTI personality model and compare them to find out the most suitable for this task. In our experiments on the MBTI Kaggle benchmark dataset, we achieved 88.63% in terms of accuracy and 88.97% of F1-Score which allow us to outperform current state-of-the-art results.