The existing frequent closed graph(FCG) mining methods under standalone mode can not process massive Internet datasets, by improving the Apriori algorithm, an iterative algorithm AMR(Apriori based on MapReduce) was proposed to mine FCGs of large scale dynamic networks based on Hadoop. First of all, the edge sets of dynamic networks were stored in the key-value table, and a serialized subgraph coding mechanism was designed to ensure the uniqueness of frequent subgraphs(FSG). Secondly, a communication mechanism was proposed to pass subgraph codes, and to ensure the accuracy of FCG, local supports in each partition were aggregated to a global one. Finally, FCGs were achieved by pruning the FSGs. AMR was used in the dynamic networks of both country and AS level Internet, and experimental results showed that FCG can accurately characterize the topology of backbone Internet, thus verifying the efficiency and effectiveness of AMR in mining FCG of large scale dynamic networks.
