Estimation of Personality Traits From Portrait Pictures Using the Five-Factor Model

• Main Authors: Marco A. Moreno-Armendariz, Carlos Alberto Duchanoy Martinez, Hiram Calvo, Miguelangel Moreno-Sotelo
• Format: Article
• Language: English
• Published: IEEE 2020-01-01
• Series: IEEE Access
• Subjects: Personality traits estimation; five-factor model; convolutional neural networks; image analysis
• Online Access: https://ieeexplore.ieee.org/document/9244051/

This work presents a model based on Deep Neural Networks for the prediction of apparent personality. It can quantify personality traits with the Five-Factor model (Big Five) from a Portrait image. In order to evaluate the effectiveness of this approach, a new corpus of 30,935 portraits with their associated personality trait was extracted from an existing resource of videos (First Impressions, ChaLearn) tagged with redundant pairwise comparisons to ensure consistency. We propose several models using Convolutional Neural Networks to automatically extract features from a portrait that are indicators of personality traits; then the models classify these characteristics into a binary class for each Big Five factor: openness to experience (O), conscientiousness (C), extraversion (E), agreeableness (A), and neuroticism (N). In addition, we experiment with feature encoding and transfer learning to enrich the representation of images with additional untagged portraits (~45,000 and ~200M), reaching a percentage of accuracy within the state of the art (albeit not directly comparable), obtaining 65.86% as a classifier averaging the 5 factors (O=61.48%, C=69.56%, E=73.23%, A=60.68%, N=64.35%). Compared to human judgment (mean accuracy of 56.66%), the model obtained higher average performance and higher accuracy in 4 of the 5 factors of the Big Five model. In addition, in comparison with the state of the art this model shows several advantages: (1) it requires only a single portrait to make the prediction, being this a non-invasive and easily accessible resource (e.g. selfies) (2) the extraction of features from the portrait is done automatically, (3) a single model performs the extraction of characteristics and classification.