Classificação de incompatibilidades cross-browser em aplicações web com design responsivo com uso de redes neurais convolucionais

• Ano de defesa: 2022
• Autor(a) principal: Santos, Danilo Alves dos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Informática; UTFPR
• Palavras-chave em Português: Software de aplicação; Aplicações Web; Redes Neurais (Computação); Application software; Web applications; Neural networks (Computer science); CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/30193

The variety of browsers available to users, which allow accessing and executing web applications, leaves room for the occurrence of display incompatibilities or execution of components present in web pages due to the different configurations between the existing rendering engines. Even with the advancement of technologies for developing responsive pages, rendering incompatibilities are still present. When these incompatibilities occur due to running in different browsers they are known as Cross-Browser Incompatibilities, or XBI. For developers, this means more time spent performing manual inspections to detect and correct these incompatibilities. Report the advances in the use of deep neural networks associated with object classification through computer vision, this work used this approach in the classification scenario of internal layout incompatibilities in responsive web applications. For this, experiments were conducted using classifiers based on Convolutional Neural Networks. Different configurations were analyzed to identify the architecture that best suited the proposed classifier. For network training, pixels generated from screenshots from datasets from previous works were used as input data. To quantify the effectiveness of the classifier with the proposed approach, 10-Fold-Cross-Validation method was used, combined with Precision, Recall and F-Score metrics, where it was verified that the proposed model presented better results of Precision and Recall in relation to to the Browserbite strategy, but it was not able to achieve similar results to the Crosscheck and Watanabe et al. (2019) strategies, used to oppose the implemented model. Through the experiment conducted, it can be said that the main factor that influenced the low efficiency of the classifier was the unbalance of the available dataset.