By using the principle of resource allocation, we propose a recommendation algorithm which is based on the partition of directed graph. The items directed graph is established by combining with the bipartite graphs network structure and resource allocation method, and is partitioned by the method of Asymmetric Nonnegative Matrix Factorization. Then we classify items by the relationship between them, set connection weights between the items and implement a recommendation from the Top-N items to the user. Experimental results show that the proposed algorithm can improve the recommendation accuracy and the recommendation diversity, and reduce the popularity of recommendation to a certain extent.