Automatic Personality Prediction using Deep Learning Based on Social Media Profile Picture and Posts

Abstract

Uploaded contents by social media users are affected by their personality, for example the profile photo they used and the posts they published. In this research, we create an automatic prediction for Twitter users' personalities through their photo profile and their tweets, comparing the result from using either of the feature and both of them. 1290 Twitter users that had taken MBTI test from 16personalities were used as the dataset. Facial feature from profile photo is obtained by using the face detection model that is combined with smile detection such that not only can we obtain the feature of the face, but also their expressions. As for the color, the feature is obtained by their color composition, which is hue, saturation, and value. For tweets, features are obtained by using a pre-trained word vector. Our result shows that image features can predict personality better than text feature and the combination of text and image features. Based on our result, we also found that a single profile picture is capable of reliably predicting personality.