The SEWASIE MAS for semantic search

Abstract

The capillary diffusion of the Internet has made available access to an overwhelming amount of data, allowing users having benefit of vast information. However, information is not really directly available: internet data are heterogeneous and spread over different places, with several duplications, and inconsistencies. The integration of such heterogeneous inconsistent data, with data reconciliation and data fusion techniques, may therefore represent a key activity enabling a more organized and semantically meaningful access to data sources. Some issues are to be solved concerning in particular the discovery and the explicit specification of the relationships between abstract data concepts and the need for data reliability in dynamic, constantly changing network. Ontologies provide a key mechanism for solving these challenges, but the web’s dynamic nature leaves open the question of how to manage them. Many solutions based on ontology creation by a mediator system have been proposed: a unified virtual view (the ontology) of the underlying data sources is obtained giving to the users a transparent access to the integrated data sources [1, 2, 3]. The centralized architecture of a mediator system presents several limitations, emphasized in the hidden web [4]: firstly, web data sources hold information according to their particular view of the matter, i.e. each of them uses a specific ontology to represent its data. Also, data sources are usually isolated, i.e. they do not share any topological information concerning the content or structure of other sources. Our proposal is to develop a network of ontology-based mediator systems, where mediators are not isolated from each other and include tools for sharing and mapping their ontologies. In this paper, we describe the use of a multi-agent architecture to achieve and manage the mediators network. The functional architecture is composed of single peers (implemented as