Annotation for the Deep web.pdf

Published by the IEEE Computer Society Information Integration on the Web Annotation for the Deep Web Siegfried Handschuh and Raphael Volz, University of Karlsruhe Steffen Staab, University of Karlsruhe and Ontoprise GmbH One of the core challenges of the Semantic Web is to create metadata by mass collaboration—by combining semantic content created by a large number of people. To attain this objective, researchers have developed several approaches1–3 to deal with creating manual or semiautomatic metadata from existing information. However, most of these approaches build on the assumption that the information sources are static, such as static HTML pages or books in a library (scenarios A and B in Table 1). Today, however, a large percentage of Web pages are dynamic. Estimates about the ratio of static to dynamic pages based on Web pages actually crawled by search engines typically conclude that dynamic Web pages outnumber static ones by 100 to 1. Man- ually annotating every single dynamic Web page generated—for example, from a database that con- tains a catalog of books—would be tedious. It would be better to annotate the database so that it is reusable for site-specific Semantic Web purposes. To achieve this objective, several approaches pro- vide for constructing wrappers by explicitly defin- ing HTML or XML queries or by learning such def- initions from examples4 (scenario C in Table 1). See the “Related Work” sidebar for a discussion of related approaches. It has become possible with these approaches to create metadata manually for a set of structurally similar Web pages. These approaches offer the advantage of not requiring the database owner’s cooperation. But the drawback to the wrap-