Personality Recognition Using Transformer Model: A Study on the Big Five Traits

Personality recognition, pivotal in artificial intelligence and computational psychology, holds promise for applications ranging from psychological diagnostics to personalized user experiences. This study proposes a transformerbased architecture, the ELECTRA model, for classifying the big five personality traits (openness, conscientiousness, extroversion, agreeableness, neuroticism) from textual data. Five independent ELECTRA classifiers were trained on the Pennebaker and King Essays dataset, each fine-tuned for a single trait. To address data scarcity, synonym replacement augmentation was employed, enriching textual diversity while preserving semantic coherence. The models achieved robust performance, with accuracies of 75% (openness), 72% (conscientiousness), 78% (extroversion), 74% (agreeableness), and 74% (neuroticism), alongside consistently strong AUC scores (>0.75), underscoring their discriminative capability. The results highlight ELECTRA's suitability for scalable personality recognition, balancing computational efficiency with accuracy. This work advances automated psychometric analysis, offering a framework adaptable to realworld scenarios like mental health screening or adaptive humancomputer interaction.