In the unstructured Peer-to-Peer systems, a core operation is efficient location of resources. Conventional informative searching algorithms, however, always cannot perform well under peer churn rate network environments. In this paper, we designed a robust unstructured peerto-peer files sharing system. When forwarding query, we usually choose the peer with the highest forward probability. But in this paper we didn't compute the forward probability according to the current neighbor node's individual search history but to a leaning model based on machine learning technologies. We proposed a more robust way for searching. The experimental results show that our methods are more effective and efficient under higher peer churn environment.
Related WorksTo overcome the shortcoming of the simple flooding algorithm, researchers presented some improved algorithms. Expanding ring algorithm uses floods with increasing TTLs which can halves the per-node message overload [2]. In the same paper, Random Walk algorithm for searching was also presented, and the experimental results showed that the 32 walker random walk can reduce