ExtraWeb: um sumarizador de documentos Web baseado em etiquetas HTML e ontologia

Abstract

This dissertation presents an automatic summarizer of Web documents based on both HTML tags and ontological knowledge. It has been derived from two independent approaches: one that focuses solely upon HTML tags, and another that focuses only on ontological knowledge. The three approaches were implemented and assessed, indicating that associating both knowledge types have a promising descriptive power for Web documents. The resulting prototype has been named ExtraWeb. The ExtraWeb system explores the HTML structure of Web documents in Portuguese and semantic information using the Yahoo ontology in Portuguese. This has been enriched with additional terms extracted from both a thesaurus, Diadorim and the Wikipedia. In a simulated Web search, ExtraWeb achieved a similar utility degree to Google one, showing its potential to signal through extracts the relevance of the retrieved documents. This has been an important issue recently. Extracts may be particularly useful as surrogates of the current descriptions provided by the existing search engines. They may even substitute the corresponding source documents. In the former case, those descriptions do not necessarily convey relevant content of the documents; in the latter, reading full documents demands a substantial overhead of Web users. In both cases, extracts may improve the search task, provided that they actually signal relevant content. So, ExtraWeb is a potential plug-in of search engines, to improve their descriptions. However, its scability and insertion in a real setting have not yet been explored.