Deep Learning Based statistical model for identification of personality traits in text

Abstract

Human personality is significantly represented by those words which people use in their newlinespeech or writing. People use various social media sites to share various kinds of personal newlineinformation and write what they feel. So, personality detection based on texts from online newlinesocial networks has attracted more and more attention. The Big Five Model is the most newlineinfluential model and is widely accepted for measuring and quantifying people s personality. newlineIt consists of five aspects: Openness, Conscientiousness, Extroversion, Agreeableness and newlineNeuroticism. newlinePersonality is a fundamental aspect of human behaviour, emotion, motivation, and thought newlinepattern characteristics. It has a great impact on our lives. Personality traits reveal a person s newlinethought, about their feelings and behaviour. Hence it portrays that psychologically one person newlineis different from another. Henceforth its prediction is the vital research area. newlineIn this research, we have reviewed significant learning algorithms (NB, EC, DTC, CNN) newlinewhich have been employed for personality traits detection. Eventually, we have used LSTM newlinebased deep neural network classifier. In pre-processing, we have used time efficient sentence newlinetokenization algorithm; and also given weight to emojis, as it also contributes to personality. newlineThe LSTM classifier consists of sequence input layer, word embedding layer, LSTM layer, newlinedropout layer, fully connected layer, softmax layer and classification layer to predict the newlineaccuracy of personality traits from essay dataset. newlineFurther, we have compared the accuracy of LSTM based deep neural network classifier with newlineNB, DTC, Ensemble and CNN classifier. We have also compared brute force tokenization newlinemethod with finite automata tokenization method over dataset of different sizes. Finally, after newlinethe practical implementation, we have achieved significant accuracy as 65%.