Machine learning para análise exploratória e evidenciação empírica da relação entre solidão, política e mídias sociais no Brasil

Abstract

Loneliness in contemporary society has emerged as a complex social problem. It is increasingly recognized as a global public health crisis due to its significant impact on populations across various countries. In Brazil, there is still a scarcity of studies on the impacts of loneliness, its correlated factors, and its demographic and economic implications, largely due to limited literature and the recent standardization of assessment tools, such as the UCLA Loneliness Scale, for evaluating this construct. This study integrates artificial intelligence methods, specifically computational statistical techniques, for the analysis of psychological data. The aim was to provide empirical evidence for the associations between behaviors related to this under-investigated issue. By training machine learning algorithms, it was possible to analyze the complex patterns of association between levels of loneliness, political behavior, and social media use within the context of the 2022 Brazilian presidential elections. Online self-report instruments were administered to a sample of 176 participants, predominantly residents of Southern Minas Gerais. Algorithms were implemented in Python within the VS Code environment. Analysis with K-means indicated that the high dimensionality of the data prevented complete clustering during pre-training. However, the analysis revealed clusters highlighting the severity of loneliness by age group, with a higher prevalence among young adults. KNN did not show significant associations between the analyzed variables, finding only a very weak relationship between loneliness severity and social media use. In contrast, DBSCAN managed to cluster the data more clearly, making it possible to identify relationships between some of the severity levels of the three variables, especially when these severity levels were low. The results indicated few cases of high severity for loneliness and political behavior in the sample. No cases of extreme severity associated with social media use were found. These latter results may be associated with the limitations of the sample and the collected data. In conclusion, this study contributes to clarifying hypotheses about the relationships between the analyzed constructs and provides a methodology that may eventually allow for the assessment of the relationship between three emerging social phenomena. Within the investigated context, the data revealed very weak associations between loneliness, political behavior, and social media use. These findings suggest the need for further studies, primarily adjustments to the database for better representativeness and scope, as well as refinements to the implemented models, considering the complexity and particularities of the data.