Mining association rules is an essential task for knowledge discovery. But discovering association rules based on graph need many times to traverse graph in generating candidate itemset. This paper proposes the improved algorithm, which constructs the directed graph and generate candidate item sets by using the directed neighbor nodes set, the algorithm need traverse the directed graph only once. The algorithm verifies whether a candidate itemset is a frequent itemset by logic AND operation. Experimental result shows that the improved algorithm has better efficiency than other algorithms based on graph.