Prediction of MBTI with textual data using different pre-trained transformer models
Shreyash Arya, Ann Maria Joy, Jeshma Nishitha Dsouza · 2023
Social media has taken over every aspect of our lives. We use it to communicate with friends and family, learn about current affairs, and express ourselves. The extent to which personal information is revealed on social media can be a very useful indicator of an individual's personality. In our research, we have used Myers Briggs Type Indicator (MBTI) to classify individuals into 16 different types. As the name suggests, MBTI is an indicator rather than a test. It is built to sort and indicate a person's personality trait based on a questionnaire, making it possible to assess their personality based on the online posts. Our aim was to conduct a comparative analysis on various pre-trained language models, namely BERT, RoBERTa, DistilBERT, and ALBERT, which represents the forefront of natural language processing and understanding. Our investigation delved into the effectiveness of these models in deciphering the significance of social media data, where character limitations and informal language are predominant. In addition, we explored their abilities to accurately predict an individual's MBTI personality type based on the textual content they share. We observed that among the models addressed, the best performance was showcased by BERT with accuracy of 74.59%.