Integrating External Sources in a Corporate Semantic Web Managed by a Multi-agent System

We first describe a multi-agent system managing a corporate memory in the form of a corporate semantic web. We then focus on a newly introduced society of agents in charge of wrapping external HTML documents that are relevant to the activities of the organization, by extracting semantic Web annotations using tailored XSLT templates. Agents and corporate semantic webs Organizations are entities living in a world with a past, a culture and inhabited by other actors; the pool of knowledge they mobilize for their activities is not bounded neither by their walls nor by their organizational structures: organizational memories may include or refer to resources external to the company (catalogs of norms, stock-markets quotations, digital libraries, etc.). Our research team currently studies the materialization of a corporate memory as a corporate semantic web; this follows the general trend to deploy organizational information systems using internet and web technologies to build intranets and intrawebs (internal webs, corporate webs). The semantic intraweb that we shall consider here, comprises an ontology (O'CoMMA [Gandon, 2001]) encoded in RDFS, descriptions of the organizational reality encoded as RDF annotations about the groups (corporate model) and the persons (user profiles), and RDF annotations about the documentary resources. The result of this approach is a heterogeneous and distributed information landscape, semantically annotated using the conceptual primitives provided by the ontology. To manage this corporate knowledge, it is interesting to rely on a software architecture that is itself heterogeneous and distributed, and the adequacy of multi-agent systems have been acknowledged in a range of projects proposing multi-agent systems addressing different aspects of knowledge management inside organizations. CASMIR [Berney and Ferneley, 1999] and Ricochet [Bothorel and Thomas, 1999] focus on the gathering of information and Copyright © 2002, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. the adaptation of interactions to the user’s preferences, learning interest to build communities and collaborative filtering inside an organization. KnowWeb [Dzbor et al., 2000] relies on mobile agents to support dynamically changing networked environment and exploits a domain model to extract concepts describing documents and use them to answer queries. RICA [Aguirre et al., 2000] maintains a shared taxonomy in which nodes are attached to documents and uses it to push suggestions to interface agents according to user profiles. FRODO [Van Elst and Abecker, 2002] is dedicated to building and maintaining distributed organizational memories with an emphasis on the management of domain ontologies. Finally, our team participated to the European project CoMMA [CoMMA, 2000], which aimed at implementing and testing a corporate memory management framework based on agent technology, for two application scenarios: (1) ease the integration of a new employee to an organization and (2) assist the technology monitoring activities. The system does not directly manage documents, but annotations about documents referenced by their URIs. CoMMA focused on three functionalities: (a) improve precision and recall, to retrieve documents using semantic annotations; (b) proactively push information using organization and user models; (c) archive newly submitted annotations. The architecture of CoMMA as it was demonstrated at the end of the project will be detailed in the following section. By annotating resources available on the open Web, the organizational memory may span the organizational boundaries. In CoMMA, some agents were in charge of assisting the annotation and archiving process, but the annotation of a resource was essentially a manual process. While it was acceptable then, for the scenarios involved people whose role included the annotation task, it is clear that tools are needed to ease this work by assisting the exploitation of structural clues in resources, by saving the user from repeating similar and tedious manipulations, by automating updates when resources change, etc. It is for this reason that we introduced a new society in CoMMA, as described in the third section. From: AAAI Technical Report SS-03-01. Compilation copyright © 2003, AAAI (www.aaai.org). All rights reserved.