Semantic Content Analysis Supporting Web Accessibility Evaluation

Abstract

Despite its many advantages, automated web accessibility evaluation's main limitation is still its inability to assess rules and techniques that require a semantic understanding of the web content. Today, machine learning solutions are available that can interpret different media content with a reasonable degree of confidence. These solutions have an untapped potential to increase the accessibility of web pages. This potential extends to the evaluation of web pages also. This paper proposes an algorithm to automatically rate the similarity between a content and its textual description in a web accessibility evaluation context. The validity of the algorithm is demonstrated by comparing its ratings of descriptions of images obtained from their alt texts with human ratings of the same descriptions.