SERSE: Searching for Digital Content in Esperonto

This paper presents SERSE, a multi-agent system that combines different technologies such as peer-to-peer, ontologies, and multi-agent technology in order to deal with the complexity of searching for digital content on the Semantic Web (SW). In SERSE, agents communicate and share responsibilities on a peer-to-peer basis. Peers are organised according to a semantic overlay network, where the neighbourhood is determined by the semantic proximity of the ontological definitions that are known to the agents. The integration of these technologies poses some problems. On the one hand, the more ontological knowledge the agents have, the better we can expect the system to perform. On the other hand, global knowledge would constitute a point of centralisation which might potentially degrade the performance of a P2P system. The paper identifies five requirements for efficiently searching SW content, and illustrates how the SERSE design addresses these requirements. The SERSE architecture is then presented, together with some experimental results that evaluate the performance of SERSE in response to changes in the size of semantic neighbourhood, ranging from strictly local knowledge (each agent knows about just one concept), to global knowledge (each agent has complete knowledge of the ontological definitions).