Semantic Web Recommender Systems 79 Ontological Commitment. Basically, the Semantic Web is made up of machine readable content distributed all over the Web. In order to ensure that agents can understand and reason about the respective information, semantic interoperability via ontologies or common content models must be established. For instance, FOAF 5], an acronym for"Friend of a Friend", defines an ontology for establishing simple social networks and represents an open standard agents can rely upon. Interaction Facilities. Decentralized recommender systems have primarily been ubject to multi-agent research projects. In suchlike settings, environment models are agent-centric, enabling agents to directly communicate with their peers and thus making synchronous message exchange feasible. The Semantic Web, being an ag gregation of distributed metadata, constitutes an inherently data-centric environment model. Messages are exchanged by publishing or updating documents encoded in RDE OWL. or similar formats. Hence the communication becomes restricted to asynchronous message exchange only Security and Credibility. Closed communities generally possess efficient means to control the users'identity and penalize malevolent behavior. Decentralized systems nong those peer-to-peer networks, open marketplaces and the Semantic Web, like- ise, cannot prevent deception and insincerity. Spoofing and identity forging thus become facile to achieve [22]. Hence, some subjective means enabling each indi- vidual to decide which peers and content to rely upon are needed. Computational Complexity and Scalability. Centralized systems allow for es- timating and limiting the community size and may thus tailor their filtering sys- tems to ensure scalability. Note that user similarity assessment, which is an integral part of collaborative filtering [6], implies some computation-intensive processes The Semantic Web will once contain millions of machine-readable homepages. omputing similarity measures for all these "individuals"thus becomes infeasible. Consequently, scalability can only be ensured when restricting these computations to sufficiently narrow neighborhoods. Intelligent filtering mechanisms are needed, ill ensuring reasonable recall, i.e., not sacrificing too many relevant, like-minded rents. Low Profile Overlap. Interest profiles are generally represented by vectors indicat ing the user's opinion for every product. In order to reduce dimensionality and ensure profile overlap, some centralized systems like Ringo [20] require users to rate small subsets of the overall product space. These mandatory assessments, provisional tools for creating overlap-ensuring profiles, imply additional efforts for prospective users Other recommenders, among those GroupLens and MovieLens [14], operate in do- mains where product sets are comparatively small. On the Semantic Web, virtually no restrictions can be imposed on agents regarding which items to rate. Hence, new approaches to ensure profile overlap are needed in order to make profile similarity measures meaningful 3 Proposed Approach Endeavors to ensure semantical interoperability through ontologies constitute the corner- stone of Semantic Web conception and have been subject to numerous research projectsSemantic Web Recommender Systems 79 – Ontological Commitment. Basically, the Semantic Web is made up of machine￾readable content distributed all over the Web. In order to ensure that agents can understand and reason about the respective information, semantic interoperability via ontologies or common content models must be established. For instance, FOAF [5], an acronym for “Friend of a Friend”, defines an ontology for establishing simple social networks and represents an open standard agents can rely upon. – Interaction Facilities. Decentralized recommender systems have primarily been subject to multi-agent research projects. In suchlike settings, environment models are agent-centric, enabling agents to directly communicate with their peers and thus making synchronous message exchange feasible. The Semantic Web, being an ag￾gregation of distributed metadata, constitutes an inherently data-centric environment model. Messages are exchanged by publishing or updating documents encoded in RDF, OWL, or similar formats. Hence, the communication becomes restricted to asynchronous message exchange only. – Security and Credibility. Closed communities generally possess efficient means to control the users’identity and penalize malevolent behavior. Decentralized systems, among those peer-to-peer networks, open marketplaces and the Semantic Web, like￾wise, cannot prevent deception and insincerity. Spoofing and identity forging thus become facile to achieve [22]. Hence, some subjective means enabling each indi￾vidual to decide which peers and content to rely upon are needed. – Computational Complexity and Scalability. Centralized systems allow for es￾timating and limiting the community size and may thus tailor their filtering sys￾tems to ensure scalability. Note that user similarity assessment, which is an integral part of collaborative filtering [6], implies some computation-intensive processes. The Semantic Web will once contain millions of machine-readable homepages. Computing similarity measures for all these “individuals” thus becomes infeasible. Consequently, scalability can only be ensured when restricting these computations to sufficiently narrow neighborhoods. Intelligent filtering mechanisms are needed, still ensuring reasonable recall, i.e., not sacrificing too many relevant, like-minded agents. – Low Profile Overlap. Interest profiles are generally represented by vectors indicat￾ing the user’s opinion for every product. In order to reduce dimensionality and ensure profile overlap, some centralized systems like Ringo [20] require users to rate small subsets of the overall product space. These mandatory assessments, provisional tools for creating overlap-ensuring profiles, imply additional efforts for prospective users. Other recommenders, among those GroupLens and MovieLens [14], operate in do￾mains where product sets are comparatively small. On the Semantic Web, virtually no restrictions can be imposed on agents regarding which items to rate. Hence, new approaches to ensure profile overlap are needed in order to make profile similarity measures meaningful. 3 Proposed Approach Endeavors to ensure semantical interoperability through ontologies constitute the corner￾stone of Semantic Web conception and have been subject to numerous research projects