Overview of Text Based Personality Prediction Using Deep Learning

Abstract

Text-Based Personality Prediction (TBPP) has garnered increasing attention in recent years, particularly within the frameworks of the Myers-Briggs Type Indicator and the Big Five Personality Model. This study presents a comprehensive systematic review of TBPP methodologies, focusing specifically on research published since 2017. Leveraging Google Scholar, a meticulous selection process was employed to identify and analyze papers meeting relevance criteria. The selected studies were analyzed for research design, data collection methods, preprocessing techniques, and modeling approaches. Notably, the study identifies prevalent Natural Language Processing methods utilized in TBPP, such as Recurrent Neural Networks, Convolutional Neural Networks, Long Short-Term Memory networks, ensemble methods, and pre-trained models like BERT. Results indicate that combining knowledge graphs with Bi-LSTM models achieved the highest accuracy for Big Five traits at 71.5%, while a BERT-CNN-RNN ensemble reached 85% accuracy for MBTI. The synthesized findings offer valuable insights into the current landscape of TBPP, with the aim of informing both researchers and practitioners. Furthermore, the study provides recommendations for future research directions, emphasizing the importance of refining methodologies and addressing challenges to foster continued innovation in personality prediction within the TBPP domain.