In view of the bipartite graph method of network-based inference (NBI) only considered whether users evaluated the project or not, but not given their scores, the thesis proposed a preferential network-based inference (PNBI) recommended method. Based on inference network, the method takes into account that user's rating values for the program reflects his degree of preference. In the "User-Item" resource allocation process, the method allocates resources to the item that gets a higher score, this method can overcome the NBI algorithm's disadvantage of failing to use low score value. Considering the sparsity of data, the method uses inverted list to discrese the number of calculation to accelerate the algorithm. Experiments on MovieLens dataset show that, PNBI bipartite graph recommended algorithm outperforms NBI bipartite graph recommended algorithm in accuracy, coverage and recall.