Personality BERT: A Transformer-Based Model for Personality Detection from Textual Data

Understanding personality type can aid in understanding people preferences and associated cognitive processes. Automated personality detection can commendably help NLP experts and psychoanalysts to identify the dominant or distinguishing qualities of a person. At its basic level, a personality is expressed through a person's temperament or emotional tone. Pertinent studies validate linguistic cues in written and spoken text as a coherent and consistent mode of assessing and interpreting personality. With the proliferation of social media applications, the psycholinguistic markers in user's online posts can facilitate comprehending variations in personalities. Transformer models have emerged as new generation NLP models and are already being implemented to benefit an array of NLP use cases. This research puts forward a transformer-based model for personality detection from textual data. The proposed personality BERT is a textual modality-specific deep neural model that fine-tunes a pretrained bidirectional representation for transformers (BERT) for the personality classification task. Kaggle's MBTI dataset is used to evaluate and validate the proposed model. An fl score of 0.6945 is reported.