We present a user profile driven framework to allow individual users to organize themselves into communities of interest (CoI) based on ontologies agreed upon by all community members. In this paper, we describe the overlay network architecture to support the basic functionalities of a Col. The basic tenet of this architecture is the use of ontologies to represent objects in order to enable semantic resource discovery and retrieval which reflect the interest of the user within a specific community. Three advertising and retrieval schemes, namely aggressive, crawler-based and minimum-cover-rule, are discussed and investigated using an emulation-based and a simulation-based experimental frame-works. The results show that the minimum-cover-rule scheme exhibits higher performance than the other two schemes in the stable environment. In the high-churn environment, however, the effectiveness of the aggressive scheme is better than those of the other two schemes.