Similar to the current Web, the key to realizing the Semantic Web is scale. Arguably, to achieve this, we need a good balance between participation cost and perceived benefit. The major obstacles lie in coping with large numbers of ontologies, authors and physical hosts, inconsistent or inaccurate statements and the large volume of instance data. Our focus is on scalability through distribution. Most current approaches split ontologies into triples and distribute them among peers participating in a structured peer-to-peer overlay. Identifying a series of drawbacks with this, we propose an alternative model where each peer maintains control of its ontologies.
It is inspired on how humans do it: if you get a question where you do not have an answer on, you try to identify one or some people of which you know they have some knowledge on the subject. Then, these persons are contacted and try to answer the question. In the model we propose todo something similar, but then to Semantic Web data. We distribute the different ontologies over a number of computers and let each computer register the most important topics from the ontologies they store in a distributed ‘yellow page’. Now, if a peer gets a question, it tries to answer by itself, and those terms it cannot reason about (and therefore not answer) are asked to other peers that registered themselves for these terms.