Research On Content-Based Information Search Mechanism In Unstructured P2P Network

With the rapid growth of Peer-to-Peer (P2P) networks,the shared data information in P2P network such as text, images, audio, video are growing rapidly.However, Peer-to-Peer systems are limited in file retrieval by using some keywords,it retrieves the shared files in the network using a set of keywords gained from Boolean arithmetic, which is difficult to adapt to network environment of P2P. Therefore it is necessary to introduce the content-based retrieval (CBR) into P2P network search mechanism or expand the centralized CBIR system to distributed P2P network.Currently, distributed system based on P2P can be classified into unstructured and structured systems. The former one is based on DHTs, the use of DHTs leads to the fact that only keyword exact-match is available, which can't support the fuzzy search and also can't adapt to content-based information retrieval. So, unstructured method is used in this paper. Combining small-world theory, Peers are divided into several groups by similarity from the considering of users'interest and shared files contents. And then, it searches in-group in order to improve search efficiency and reduce the consumption of bandwidth.Firstly, the strategy of dividing groups based on users'interest is studied whose advantages and shortcomings are analyzed. Then a reformative search mechanism is proposed. In this mechanism historical feedback information is used to adjust the division of groups dynamically, which can reflect the variety of shared contents in each peer and can attain balance among management, communication cost and search efficiency.Secondly, a novel search model based on class clustering is proposed. The basic idea of the search mechanism is to link the peers that have the congener shared files. Then a cluster is formed, all the peers in a cluster have the same class of files, so it is also called class cluster. When a peer is added into network, it should find the corresponding class cluster according to all kinds of information to trigger query. The convenient links among class clusters are established to locate the target cluster efficiently by shorting the searching path. Then the routing table is looked up for the query content and forwarding direction is chosen, the target cluster is located quickly. Each cluster contains the files satisfying the query, so broadcasting inside the cluster can search it. The routing table is updated in order to get newest routing information. Simulation...