Reducing Semantic Heterogeneity of Unstructured P2P Systems Through Gossip-Based Ontology-Driven Topology Adaptation

We present GoOD-TA, a gossip-based, ontology-driven topology adaptation protocol that reduces the semantic heterogeneity generated by the use of several ontologies within P2P information sharing systems. Peers regularly exchange other peers' descriptors in order to choose neighbours that are semantically close in terms of used ontology or ability to translate queries. In GoOD-TA, the descriptors are expressive enough to define a relevant proximity function between peers. Several mechanisms enable to reduce the volume of transfered data between peers and to manage the dynamicity of the system. Particular attention has been paid to handle the evolution of peers semantic knowledge, when they change their ontology or the correspondences they know. We implemented the GoOD-TA protocol in Java, and we used the PeerSim simulator to evaluate it. We used a set of real-world ontologies that are actively used in biomedical communities. Results show that the GoOD-TA protocol allows to reduce the semantic heterogeneity related to the system topology. We also show that it handles the evolution of peers semantic knowledge, and that it is suitable for highly dynamic systems. In addition to its nice behavior, unlike other works, GoOD-TA does not require any explicit description of the dynamic overlays that should be maintained during the system lifetime