Text-Based Personality Recognition Based on User Content

Abstract

Personality recognition from textual data, a topic of growing interest, has gained significant importance in the fields of psychology, marketing, and human-computer interaction. This research explores the domain of text-based personality recognition, focusing on user-generated content to uncover the complex aspects of an individual's personality traits. It leverages state-of-the-art transformer-based models, including BigBird, Albert, and DistilBERT, enhanced with NLP statistical features. The primary objective is to evaluate and compare these cutting-edge models' performance comprehensively, concatenated with NLP statistical features, against conventional methods for personality trait recognition across diverse textual datasets, including Facebook and essay datasets. This research employs two classifiers, BiGRU and BiLSTM, to classify the five personality traits of the Big Five personality trait model using Facebook and essay datasets. BigBird, when combined with NLP statistical features and using BiLSTM as a classifier, achieves impressive F1 scores for traits EXT, NEU, AGR, CON, and OPN, demonstrating accuracies that underline the effectiveness of this approach. The findings show that pre-trained models in combination with NLP statistical features have improved the performance of the Personality recognition model in terms of accuracy and F1-score across the myPersonality datasets.BigBird