A New Approach for Peer-to-Peer Information Retrieval Systems

摘要

Up to now searching in Peer-to-Peer information retrieval systems (P2P-IR) has been limited due to the structure used or has been very complex according to the administration effort. The main principles of existing systems are: either to distribute documents to other peers, even several times, or to assign peers to fixed clusters. In the first case, the characteristics of the peers get lost and in the second case the search is limited because of the cluster boundaries. You had to decide whether to get a fast but incomplete search or a complex system with several overlays. In this paper a novel P2P-IR-System with a new principle for searching is proposed. Therefore the disadvantages of existing structured and unstructured systems and search mechanisms are considered. The new system has to guarantee a complete search with minimized complexity. To facilitate this, the peers use bloom filter as addresses and form a global, dynamical and incomplete grid-system. All peers together establish a smart, global cluster with soft and dynamic boundaries between the topics. The number of Peer-to-Peer overlays is reduced to a minimum to decrease complexity. The positioning of the peers is based on the peer characteristics and the grid is self-balanced due to this positioning. Also the search follows the same principles of the positioning and is based on similarity. Therefore no additional effort for searching is needed and all possible results are found on searching.