Deep Web Annotation Using Goal-Oriented Special Purpose Ontologies (Position Paper)

Current efforts in the semantic web area have not yet led to applications that deliver additional real-life value for web users. The main reason is a vicious circle of low populated concepts, low value for web users as well as missing incentives for annotating data. As a solution a pragmatic approach based on goal-oriented special purpose ontologies derived from the deep web in combination with annotation of instances in the deep web is presented. Finally, a use case is underpinning the idea. Introduction and Motivation Evolving semantic web technologies as well as first semantic web applications, bring forward the vision of a more powerful semantic web. However, real-life value for web users is still lacking. General purpose ontologies such as FOAF, SIOC and others do not yet generate a real addon in comparison to typical web 2.0 pages. Also semantic knowledge bases such as OpenCyc, DBpedia or OpenCalais currently don’t exceed the value of nonsemantic knowledge bases such as Wikipedia. Two reasons account for these facts: First, many concepts but only few instances are semantically described. Imagine a food retail consumer having the goal of creating his weekly shopping list. Concepts such as food item or shopping center don’t really help unless instances of food items and shopping centers are attached to those concepts. Right now, there are no incentives for data owners to annotate their data, since web users don’t use semantic applications. This will last as long as no value is generated by semantic web applications and no value is generated unless instance data is annotated – ending in a vicious circle. The second reason for limited value of current semantic web applications is the lack of special purpose ontologies and is related to the first reason. In order to break the Copyright © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. vicious circle described above, it is necessary to build semantic web applications that generate value, i.e. enable web users to achieve certain goals with a better performance than using traditional web applications. General purpose ontologies promote a top-down approach for building such applications with the promise of providing a solution for a broad range of goals within different domains. However, the effort for generating a critical mass of annotated instances increases with this approach, ending in web applications with low chances of generating value for a web user. Goal-Oriented Deep Web Annotation In this position paper a pragmatic research framework is presente, for generating limited scope but high value semantic web applications. Instead of using a top-down approach, a bottom up approach is suggested by annotating existing instances to ontologies that are build based on existing data structures on the web. It is estimated that in 2002 the web of fixed web pages (surface web) consisted of 167 TB of data whereas the database driven websites that create web pages on demand (deep web) comprised 91,850 TB of data (Lyman & Varian 2003). By retrieving the underlying concepts as well as the underlying data structures of the deep web, special purpose ontologies can be derived. Based on these ontologies, corresponding instances of the deep web can be retrieved and annotated, leading to highly populated concepts. Highly populated concepts lead to a high TermRank (Ding et al. 2005) making it easy to find the concept (e.g. with semantic search engines such as Swoogle) and increasing reuse of the concept. A similar approach has been proposed and studied earlier (Handschuh et al. 2003a, Handschuh et al. 2003b, Volz et al. 2004). However, in addition to these studies it is suggested that in order to create semantic web applications with high real-life value, the deep web should be annotated in a specific enough, human-centered domain. The NAICS