Automatic Classification of Stigmatizing Articles of Mental Illness: The Case of Portuguese Online Newspapers.

The stigma related to mental health continues to be present in online newspapers, where mental diseases are often used metaphorically to refer to entities or situations outside the clinical of mental health. This project explores the implementation of Artificial Intelligence and Natural Language Processing techniques for the task of automatically classifying stigmatizing articles with references to the mental disorders of schizophrenia and psychosis. This work is implemented in Portuguese online news articles, collected from the Arquivo.pt repository, a public repository of archived Portuguese web pages, and can be adapted to other languages or similar problems. Nine machine and deep learning algorithms were implemented, most of them yielding results with a precision above 90%. In addition, the automatic detection of articles topics was also performed, through topic modeling with the top2vec model, which allowed concluding that the stigmatization of mental health occurs, essentially, in Economics and Politics related news. The results confirm the existence of stigma in Portuguese newspapers (52% of the 978 articles collected) and the effectiveness of the use of Artificial Intelligence models to detect it. Additionally, a set of 978 articles collected and manually classified with the classes ["stigmatizing" , "literal"] is obtained.