A Research On Personalized Resource Discovery Technology Based On Learning For P2P Network

As a rising network computing mode, Peer-to-Peer (P2P) network attracts more and more attentions in the both academe and industry, and becomes a research hotspot in the computer field. Unlike the traditional Client/Server network mode, the p2p mode doesn't need to rely on the support of centralized servers; peers can freely share resources with each other by creating directly connections. Because of having avoided the bottle-neck problems in the server end point, the p2p network gets widely used in many fields such as file sharing field and so on. And as the increasing application of p2p network, how to search the resources efficiently has became a key problem in p2p network.In this thesis, we analyzed the search mechanism of unstructured p2p networks at first. And then proposed an improved p2p search model to resolve the low search efficiency, more redundant messages, and the blind routing problems in unstructured p2p networks. In the model we proposed, it makes use of the module's collaboration with each other to accomplish the model's resource discovery process. And in the document representation module of this model, it uses the vector space model to preprocess the shared document resources by nodes, so that it can form the feature of documents. And then using the routing module to record the relationships between query and response nodes, in order to do an effective routing forward to the succeeding similar queries, and avoid some uncorrelated nodes getting the query messages. And then using the friend management module to select those nodes that have high response rates as its neighbor nodes to form the node's interest group. By creating the node's interest group, we periodically shorten the distance between the query node and the high response ratio node. So as to respond the query message in a lesser hop and reduce the search depth,Finally, we evaluated this model by simulation. The simulation result shows that, compared to the flooding method, this model has obvious improvement in search success rate, the number of visited nodes and the total messages.