Abstract: This paper proposes a new link prediction model basing on community partition to differentiate diverse contribution of different vertexes which taking Common Neighbor algorithms as examples.Firstly,the model uses Clauset-Newman-Moore algorithm to partition community and generates pairs via a different aspect,then applies different weight to inner-community and outer-community neighbor vertexes,finally provides a supervised training method to infer the missing links. The experimental results demonstrate that the algorithm can not only raises prediction precision but also provides a kind of effect method for research of similarity link predication.