Classificação automática de widgets e seus subcomponentes por meio de um pipeline de aprendizado de máquina atuando em registros de mutações do DOM

• Main Author: Rizo, Eduardo Henrique
• Publication Date: 2019
• Format: Master thesis
• Language: por
• Source: Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))
• Download full: http://repositorio.utfpr.edu.br/jspui/handle/1/5299
• topic: Aprendizado do computador; Classificação; Projeto de acessibilidade; Machine learning; Classification; Barrier-free design; CNPQ::CIENCIAS EXATAS E DA TERRA; Ciência Da Computação

Since the emergence of Web 2.0 advent and the Asynchronous JavaScript and XML (AJAX) movement, web site developers have stepped up the use of sophisticated interaction mechanisms, called widgets, to design Rich Internet Applications (RIA) user interfaces. Despite the fact improves the usability and navigability of web sites, many widgets are currently implemented without the accessibility design solutions standardized in the Accessible Rich Internet Applications (ARIA) specification and hence they may be not accessible to disabled people. Notwithstanding the ARIA standardization, current development tools lack support in providing automated functions to detect nonconformities of widgets accessibility to the ARIA rules, which also contribute to its little use. Therefore, this master’s dissertation presents an approach to automatically classifying widgets type of dropdown menu, suggestion box and their respective subcomponents, through a machine learning pipeline that analyzes the changes that occur in the Document Object Model (DOM) structure of the web pages. Classifying widgets and their subcomponents is an essential step for automatic evaluation of ARIA conformance and HTML code adaptation to mitigate accessibility issues, contributing to the Software Engineering process of RIAs and keeping them in line with ARIA specifications. The proposal validation was performed by an experimental study with 34 of the 50 most visited web sites in the USA to evaluate the effectiveness of the proposed machine learning pipeline. The results provide evidence that the proposed approach is able to classify widgets with dropdown menu, suggestion box and their subcomponents with F-measure metric, that is a kind of accuracy regarding the sensitivity of machine learning classifiers, about 0.967 and 0.894 respectively. The results also suggest that the development of software artifacts that automatically identify widgets and their subcomponents, can provide subsidies for automatic accessibility evaluation tools in conformance with ARIA rules, as well as tools for automatic HTML code adaptation for accessibility, contributing to the process of web engineering accessible applications.