Predicting Personality from Social Media Text Using Multilingual BERT (mBERT)

The rapid growth of social media platforms has generated an enormous volume of user-generated textual data that reflects individuals' opinions, emotions, communication styles, and behavioral patterns. These digital footprints provide valuable information for automatically identifying personality traits using artificial intelligence and natural language processing techniques. Personality prediction has attracted significant attention due to its applications in personalized recommendation systems, adaptive learning, recruitment, digital marketing, mental health assessment, and human–computer interaction. Although numerous studies have employed machine learning and deep learning methods for personality prediction, many existing approaches rely on handcrafted features or focus primarily on English-language datasets, limiting their applicability to multilingual and low-resource languages. This paper presents a personality prediction framework based on Multilingual Bidirectional Encoder Representations from Transformers (mBERT) for analyzing social media text. The proposed approach utilizes the pre-trained mBERT model to generate contextual multilingual embeddings that capture semantic and syntactic information from user-generated posts. The extracted representations are fine-tuned for the classification of the Big Five personality traits, namely Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. The framework includes text preprocessing, tokenization, contextual embedding generation, model fine-tuning, and personality classification. Experimental evaluation is performed using standard performance metrics, including Accuracy, Precision, Recall, F1-score, and ROC-AUC, and the proposed model is compared with conventional deep learning approaches such as SVM,CNN, LSTM and BERT. The results demonstrate that multilingual contextual representations learned by mBERT effectively capture personality-related linguistic patterns and improve prediction performance across multilingual social media data. The proposed framework provides a scalable and effective solution for automatic personality prediction and offers a strong foundation for future multilingual and low-resource language personality computing research.