Evaluation of Automatic Discovery of Visual Elements of Web Pages

Web pages are typically designed for visual interaction – they include many visual elements to guide the reader. However, when they are accessed in alternative forms such as in audio, these visual elements are not available and therefore they become inaccessible. To address this problem, we have proposed an approach to identify visual elements in a web page and then characterize the semantic role of these web elements. The purpose of this technical report is to discuss the evaluation methodology in detail and present our results, which provide useful information into later applications of web page segmentation, heuristic role detec- tion of web elements and web page transcoding. Our evaluation shows that, our segmentation algorithm has success rate above average and generally, users prefer the detailed segmentation levels to the simplistic segmentation levels. Moreover, our user evaluation on role detection shows that our proposed approach has around 80% receptive accuracy, but the proposed knowledge base could be further improved for better results.