Loading...
Publication
Estudo exploratório sobre aplicações de Modelos de Linguagem de Larga Escala em acessibilidade web
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.9 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O presente documento investiga a aplicação de Large Language Models (LLMs),
especificamente do Google Gemini 2.0 Flash, na remediação de problemas de acessibilidade
web identificados pela ferramenta AXE, com foco nas diretrizes WCAG 2.2.
Com base em resultados promissores evidenciados em estudos anteriores, o potencial dos LLMs
é reconhecido como agentes facilitadores na promoção da acessibilidade digital. Este estudo
analisa, de forma empírica, a eficácia dos LLMs na resolução de quatro áreas críticas: contraste
de cor, atributos ARIA, ausência de texto alternativo/geração de descrições para conteúdo não
textual, e correção de hiperligações sem texto discernível, sendo a avaliação conduzida a partir
da análise e remediação semiautomática dos 100 websites mais populares do mundo.
Ao reunir dados quantitativos e qualitativos sobre o comportamento do modelo perante
diferentes desafios de acessibilidade, este trabalho procura ainda contribuir para o debate
sobre a viabilidade do uso de LLMs como ferramentas de apoio à acessibilidade web,
explorando tanto o seu potencial como as suas limitações práticas.
This paper investigates the application of Large Language Models (LLMs), specifically Google Gemini 2.0 Flash, in the remediation of web accessibility issues with a focus on WCAG 2.2 guidelines, identified by the AXE tool. Based on promising results demonstrated in previous studies, the potential of LLMs is recognized as facilitating agents in the promotion of digital accessibility. This study empirically analyses the effectiveness of LLMs in resolving four critical areas: colour contrast, ARIA attributes, lack of alternative text/generation of descriptions for non-text content, and correction of hyperlinks without discernible text, with the evaluation being conducted based on the analysis and semi-automatic remediation of the 100 most popular websites in the world. By gathering quantitative and qualitative data on the behaviour of the model regarding different accessibility challenges, this work seeks to contribute to the debate on the feasibility of using LLMs as tools to support web accessibility, exploring both their potential and their practical limitations.
This paper investigates the application of Large Language Models (LLMs), specifically Google Gemini 2.0 Flash, in the remediation of web accessibility issues with a focus on WCAG 2.2 guidelines, identified by the AXE tool. Based on promising results demonstrated in previous studies, the potential of LLMs is recognized as facilitating agents in the promotion of digital accessibility. This study empirically analyses the effectiveness of LLMs in resolving four critical areas: colour contrast, ARIA attributes, lack of alternative text/generation of descriptions for non-text content, and correction of hyperlinks without discernible text, with the evaluation being conducted based on the analysis and semi-automatic remediation of the 100 most popular websites in the world. By gathering quantitative and qualitative data on the behaviour of the model regarding different accessibility challenges, this work seeks to contribute to the debate on the feasibility of using LLMs as tools to support web accessibility, exploring both their potential and their practical limitations.
Description
Keywords
Large Language Models Web Accessibility WCAG 2 2 Acessibilidade web