The problem of file pollution is prevalent in P2P file sharing system nowadays. The pollution reduces the availability of system dramatically. As the same as immune system, P2P file sharing system is also highly distributive, self-adaptive and self-organized. Artificial immunology based anti-pollution object reputation mechanism is established to select neighbor peer set and improve the system availability. Vector space similarity is computed to weight the votes collected from other peers. Horting graph-theoretic approach is employed to deal with the data sparsity problem. Selfadaptive reputation threshold scheme is adopted to estimate the file authenticity. Simulation results show that the system can distinguish between good and polluted files accurately and thwart the proliferation of pollution in the network with low communication cost.