Emognizer: aplicação baseada em inteligência artificial para análise emocional de redes sociais

Abstract

Bearing in mind that a large part of the Brazilian population is active on social networks and the emotions propagated by people are a reflection of their mental health, a system was developed that detects emotions in texts produced by an individual in their Twitter and Reddit profiles. Then, it presents them through representations that provide finding signs of mental disorders. The representations are present in an application for mobile devices, built using a template that preaches for immersion and good user experience. Therefore, a survey of mental disorders that had characteristics that could be represented computationally was carried out, which resulted in four disorders: Disruptive Mood Dysregulation Disorder (DMDD), Generalized Anxiety Disorder (GAD), Specific Phobia, and Major Depressive Disorder (MDD). Also, five datasets composed of records labeled by specialists were used, which were pre-processed using Natural Language Processing (NLP) techniques and later used to carry out the training and evaluation of an LSTM (Long Short-Term Memory) Neural Network model who, during the evaluation process, showed to be able to recognize emotions in texts taken from both social networks mentioned above.