ExtraWeb - An Extrinsic Task-oriented Evaluation of Webpage Extracts

Abstract

This paper focuses on the usefulness of extracts of webpages in Brazilian Portuguese as the means to filter information for Web users to quickly and consistently judge the relevance of search engine results. ExtraWeb, an ontology-and HTML-based summarizer, has been built aiming at providing an alternative to query-biased extracts typically made available by Web search engines. An extrinsic evaluation of ExtraWeb was carried out under a controlled experiment that retrieves webpages in Portuguese and generates a set of extracts for an Internet user to evaluate. Only the relevance judgment of the extracts was assessed. Results show that the system is promising in helping users to filter relevant webpages.