DATA-DRIVEN DEVELOPMENT OF PERSONALITY PREDICTIVE LEXICA FROM SOCIAL MEDIA
Monday, March 23, 2020, 01:00pm - 03:00pm
Speaker: Xiaoli He
Location : Remote
Gerard De Melo (Advisor), Manish Singh, Yongfeng Zhang
Event Type: Masters Defense
Abstract: Automatic personality prediction is getting more popular because it is convenient and reliable. Lexicon-based analysis has been successful in the fields of sentiment analysis and emotion. Many studies have used linear models for personality prediction, which suggests that we can also use lexical-based analysis for personality prediction. In the current study, we developed weighted word lexicons (words and scores) on each dimension of MBTI personality. The lex- icons are built based on eight MBTI datasets, different features (unigram, 1-2 grams, 1-2-3 grams) and weighitings (TF, TF-IDF, TF-logIDF), and different supervised learning models. Then we ran correlation analysis between our MBTI lexicons and other existing lexicons, such as Big-5, emotion, sentiment, age, gender. The correlation analysis shows interesting and rea- sonable correlation between different personality dimensions and other psychological traits, and it also provides evidence for the robustness of our lexicons.