Understanding Personality through Social Media

In this paper, we study the relationship between language use on Twitter and personality traits. Specifically, we want to know how various linguistic features correlate with each personality trait and to what extent can we predict personality traits from language. We gather personality data from Myers-Briggs Type Indicator (MBTI) personality test which contains thinking, feeling, sensation, intuition, introversion, extroversion, judging and perceiving. Using the 90K users in our dataset, we collect most recent tweets from them and design three categories of feature, namely bag of n-grams, Twitter POS tags, and word vectors to explore the most related linguistic features for different personality traits. Analysis of these features provide insights of language use for different personalities. For instances, extroverts tend to use hashtag and phrases like ”so proud”, ”so excited”, and ”can’t wait”. People who like to use emoticon are more likely to be Sensing and Feeling personality type. Moreover, we investigate the predictive power of individual features and combined features in our analysis. With the concatenation of all the features we extracted, we can predict the personality traits with an average AUC of 0.661.