Recurrent Deep Learning to Predict Personality Types from Social Media Posts

The analysis of personality behavior has become important in many applications such as educational hypermedia systems, psychotherapy, and business recommendation systems. With the development of social media platforms (SMPs), people can express their feelings and opinions through their personal posts, comments and/or tweets. This study aims to analyze texts on SMPs to identify users' personalities according to the Myers-Briggs Type Indicator (MBTI) model. It also sought to enhance prediction accuracy and compare the findings with various deep learning techniques and previous literature. This study analyzes personality through a dataset that includes 8667 participants. Two deep learning techniques are implemented namely, long short-term memory (LSTM) and convolutional neural network (CNN). Unlike earlier literature, this research helps improve the prediction accuracy of users' personalities by building a multi-layer deep learning network and integrating the random search optimization approach. The results suggest that LSTM with the application of random search optimization outperformed previous findings. The average prediction accuracy of the four dimensions namely, Introversion-Extroversion (IE), Feeling-Thinking (FT), Intuition-Sensing (NS), and Judging-Perceiving (JP) is 85.02%.