Abstract:

Hierarchical peer-to-peer networks with multiple directory services are an appealing approach to providing federated search over large-scale networks of text-based digital libraries. However, their effectiveness and efficiency is strongly affected by the topology of the peer-to-peer network. A randomly generated topology with an arbitrary content distribution is easy to create and maintain incrementally but sub-optimal. In this paper we propose an approach to dynamic content-based topology construction that enables a controlled distribution of contents in the network by keeping similar contents near to one another and linking dissimilar contents with short paths. Our approach constructs the topology of a hierarchical peer-to-peer network efficiently in an autonomous and decentralized manner. Experiments show that the constructed topology enables effective and efficient federated search.