Guilherme BITTENCOURT
Departamento de Automação e Sistemas, Universidade Federal de Santa Catarina
Florianópolis SC Brasil
http://www.das.ufsc.br/~gb
Résumé :
Enterprise In the Web, extractor agents process classes of pages (like 'call for papers' pages, 'researchers' pages, etc) neglecting the relevant fact that some of them are interrelated forming clusters (e.g., science). We propose an architecture for cognitive multi-agent systems to retrieve and classify pages from these clusters and also extract data from it. To enable agents' cooperation, two design requirements are crucial. One is a Web vision coupling a vision for contents (classes and attributes to be extracted) to a functional vision (the role of pages in information presentation, i.e., whether a page is a list, message, class instance or garbage). The other consists of explicit representation of agents' knowledge and abilities in the form of ontologies, both about the cluster's domain and agents'tasks. Agents exchange knowledge and suggestions of links and URLs which represent interrelations among page classes to accelerate the retrieval of useful pages. Experiments with two agents responsible for the classes of scientific events and articles showed quite promising results. Functional categorization was profitable for retrieval and a detailed ontology of the cluster assured good classification performances.FREITAS, F.L.G. de; BITTENCOURT, G. An Ontology-based Architecture for Cooperative Information Agents. Proceedings of the 18th International Joint Conference on Artificial Intelligence (IJCAI'03), Morgan Kaufmann (ISBN 0-127-05661-0), pp. 37-42, Acapulco, México, August 9-15, 2003.
