Detection of Depressive Symptomatology in Written Narratives in Spanish Using Machine Learning and Semantic Ontology

This exploratory research followed a mixed methods design, in which a text analysis study was implemented with machine learning techniques, semantic ontology of depressive symptomatology and agile methodology in software development. The sample consisted of 389 young people between 18 and 30 years old (64.3% women, 35.7% men) with an average age of 22.04 years (SD = 3.51). There were six phases: 1) Preparation and validation of open questions, 2) Collection of information, 3) Adaptation and implementation of the ontology in Protégé and Python software, 4) Preparation of a control dataset, 5) Generation of the automatic method in the notebook and 6) Content rating by expert reviewers in psychology. The results show that in a sample of 50 young people between the ages of 18 and 30, the best classifier was the Naive Bayes algorithm, with an accuracy of 86%, precision of 91%, completeness of 80% and an F1 value of 72.8%.