Detecção de postagens com informações falsas sobre a pandemia do Covid-19 na rede social Instagram

• Ano de defesa: 2021
• Autor(a) principal: Cabral, Mateus Oliveira
• Orientador(a): Ciferri, Ricardo Rodrigues
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• Palavras-chave em Português: Conteúdo falso; Redes sociais
• Palavras-chave em Inglês: Instagram; COVID-19; Social network; Fake news
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/15074

This dissertation addresses the detection of false information on Instagram, the social network that has been growing more and more compared to other social media platforms. Because it is a social network with multimedia content (image, video and text), but with an emphasis on posting photos, there are few scientific research on the impacts of posts with false information that this network provides on society. This happens mainly in times of political elections or in historical events, when there is a great demand for information. Therefore, this Master’s research had as its domain the health area, with emphasis on the subject of the COVID-19 pandemic, a subject of extreme importance and big social impact. Many studies address various techniques for identifying fake news articles and/or fake posts on social networks such as Facebook, Twitter, Youtube and Whatsapp.Some studies focus on the content of the news, other studies focus on the social context through information from social networks that involves sentiment analysis, while for other studies the focus is on the temporal, which is also very much analyzed on the dynamics of posts on the social network. In this Master’s research, the source chosen to extract study data has a functional dynamic that is completely different from other social networks. Sharing the phenomena that impact the dispersion of news on social media does not work in the same way on Instagram. In addition, the posted images may contain text within the images, which creates the need to use Optical Character Recognition (OCR) based tools to extract the texts, and only then compare the extracted information in posts in Portuguese to classify whether it is false or true information. Another problem, in addition to the lack of research on false information related to Instagram, is the existence of few content datasets in Portuguese for analysis and benchmark of false information detection models, especially those containing images. The aim of this Master’s research was to investigate the detection of posts in Portuguese with false information about the COVID-19 pandemic on the Instagram social network. In this sense, the research resulted in the proposal of a machine learning model that allows the detection of false information. In addition, this research performed the compilation of a dataset related to COVID-19 to be made available for future investigations into fake content on the Instagram social network. The model was validated through experimental tests with real data. The results showed an accuracy between 96% and 99% in detecting posts with false information about COVID-19.