Abstract. The present thesis examines unstructured peer-to-peer systems in which peers share structured data. Every peer owns a local database and mappings between the schema of its database and the ones of the peers that are directly linked to it. Peers express their queries and rewrite incoming queries on their local schema. However, the available rewriting algorithms, which consider only queries that can be completely rewritten to a target schema, are not enough for an environment where peers seek, and are satisfied with, information semantically similar, but not necessarily identical, to their requests. A major problem in such systems is that peers cannot easily judge the semantic relativeness of their interests to interests of other peers. Moreover, peers cannot evaluate the semantic relativeness of answers that they receive to the queries they pose to the network.
The purpose of the thesis is the development of a system which is capable of evaluating the semantic relativeness between peer schemas, as well as between queries and their rewritten versions on the peers. To this end, we assume the existence of a shared ontology describing the domain of interest of the peers. The ontology is used to semantically annotate each peer schema, making explicit the type of information provided by it. The semantic similarity measure on which the system is based, adopts the notions of recall and precision from the field of Information Retrieval and takes into consideration the semantic annotations of the peers, the pairwise mappings between the peers and the particular queries posed. Several algorithms that compute the semantic similarity of the peers were implemented with respect to these parameters. The implementation was based on open source tools like Jena and Pellet. Data from the movie database IMDb were used for the experimental evaluation of the system.