TREAD: um framework para o reconhecimento de características textuais de postagens depressivas e ansiosas em redes sociais online

Abstract

Mental disorders, such as anxiety and depression, are classified by the WHO as social problems that affect the way people think, act, and live, affecting different areas of life and affecting physical and mental well-being. As one in eight people suffer from mental disorders, it is estimated that in 2024, disorders will affect around 970 million people. The WHO estimates that 4% of people in the world suffer from some anxiety disorder, and 5% of adults suffer from depression. In this context, the area of Affective Computing has provided different studies that seek to classify these disorders through the massive analysis of social network data and the application of machine learning models. However, it is necessary to investigate whether the content available online, such as textual posts, is capable of reflecting characteristics and markers that denote indicators of mental health, especially if there is a significant differentiation between mental health disorders, such as anxiety and depression. Therefore, this work presents TREAD: A Framework for Textual Recognition of Emotional and Contextual Characteristics Related to anxiety and depression in online social media posts. To this end, the framework involves the steps of (i) collecting data from posts in online social networking communities, (ii) data processing, (iii) extracting features about the context, emotions, and feelings with EMPATH and VADER, (iv) evaluating sequential patterns through the application of the FP Growth association rules algorithm, extraction and finally, evaluating the differentiation between patterns through clustering using algorithms such as Agglomerative Clustering. Considering the data resulting from the application of the association algorithm, the confidence of the results was above 80%, and the LIFT values were greater than 1, which indicates that the strength of the association between the items is relevant. Furthermore, considering the clustering algorithms, Agglomerative Clustering obtained 81% accuracy and F1-Score. Finally, the experimental data show the importance of social networks in this context, as well as providing a reflection on the linguistic and emotional markers that differentiate the anxiety and depression communities, serving as support for future and possible digital interventions.