Information systems are currently characterized by a continuous increase of the amount of available data, and also of the users that express queries, which may be complex, on those data.
Their is classically two main kind of approaches for the long-duration interrogation domain (which contains continuous queries in databases, stream processing and publish/subscribe): in centralized approaches, one or some computers calculate all the queries to send to the users directly the asked results, which need a powerful installation and do not let the users control their own queries. Peer-to-peer approaches, in which the users do an important part of the work, do not cause those problems; but are hard to setup, regarding to the high number of participants.
In order to avoid structural problems from centralized approaches, and also to limit the difficulties from peer-to-peer approaches, we propose a collaborative system, that rely on the mathematical relations between the different queries. Our proposition is to group the users into communities of common interests, which allow to efficiently share the processing load between the participants while highly decreasing the setup complexity.
The thesis objectives are, on the first hand, to design the algorithms and specifications enabling the implementation of such a system, and on the other hand to study the implied technical and social characteristics.