Apparent Personality Analysis Based on Aggregation Model

Apparent traits personality analysis based on multimodal traits from text, handwriting, images, and audio is a challenging problem in computer vision, signal processing, and deep learning. To improve models based just on one of the input parameters we will combine all four of them with the aggregation layer. Models based on handwriting and images are the best predictors of the Neo-PI-R results/profile derived from the results of the NEO-PI-R. In addition, to get the best results we showed different aggregation layers (Max, Min, Median, Mean). We obtain the highest prediction certainty for Consciousness, Extraversion, Agreeableness, and Neuroticism. While Openness to experience was very hard to predict with the use of the aggregation model. As the five traits are not homogenous, but consist of facets that do not necessarily converge, and deeper analysis of the facets shows that the score on the main trait is nothing more than the mean of the facets scores, and that limitation could be overcome by analyzing the facets’ behavior and predictability. This limitation can be overcome with further research done in the domain of apparent personality analysis with traits and their facets.