On the assessment of personality traits by using text mining techniques

• Autores: Montenegro, Luis; Sapino, Maximiliano; Ferretti, Edgardo; Cagnina, Leticia
• Año de publicación: 2023
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; Personality Traits; Big Five Factors; Knowledge Discovery in Databases; Text Mining; Data Augmentation
• Institución: Universidad Nacional de La Plata

This paper reports a complete experience of the Knowledge Discovery in Databases process to solve a personality trait assessment problem using text mining techniques. Given that this work is part of an interdisciplinary study between researchers from the fields of Computer Science and Psychology, in this first approach, four simple predictive algorithms were evaluated; namely: Multinomial Naive Bayes, Logistic Regression, Support Vector Machines and Decision Trees. Moreover, given the nature of the problem faced, where one person may present more than one personality trait, but not necessary all of them, it was modeled by three different classification tasks: viz. binary, multiclass and multilabel. Besides, data augmentation was used as a useful technique to improve the performance of all the classification approaches evaluated. Particularly, binary classification was the approach which took more advantage of using this technique by improving its performance on average by 13% compared to the original dataset. For three out of the five personality traits studied, it achieves weighted-F1 scores above 0.75 and in particular the highest score of 0.88 was achieved for the Responsibility trait.
Red de Universidades con Carreras en Informática