Ensemble Pre-trained Weight Deep Neural Network For Personality Classification on Social Media Posts

Personality is shown by the way a person behaves and by human characteristics, and it is relevant to people’s lifestyle, preference, and decision making. The advantages of personality understanding have many effects and make a great improvement in wide areas, such as recommender systems and human resource management. One’s personality can influence people on social media a lot, especially in this social media era, people express their responses, opinions, and feelings through posts on various social media platforms. The posts that are represented in linguistic words from each social media platform are the most among them. In this paper, we proposed the Pre-trained Weight Fast LSTM Deep Neural Network model to classified the user’s personality based on the Myers-Briggs Type Indicator (MBTI). The MBTI attempts to assign four categories of personality by a two-sided continuum: Introvert or Extrovert (I/E), Intuition or Sensing (N/ S), Feeling or Thinking(F/T) and Judgment or Perception (J/P), with a resulting 16 possible psychological types. Accordingly, we proposed an ensemble method that combines the predictions from multiple models, which are CUDA Deep Neural Network library (cuDNN) with two different pre-trained weighted vector representations and XGBoost. Eventually, our testing results outperformed the other methods, especially on class I/E, class N/S, and class F/T.