Now days, finding the association rule from large number of item-set become very popular issue in the field of data mining. To determine the association rule researchers implemented a lot of algorithms and techniques. FPGrowth is a very fast algorithm for finding frequent item-set. This paper, give us a new idea in this field. It replaces the role of frequent item-set to frequent sub graph discovery. It uses the processing of datasets and describes modified FP-algorithm for sub-graph discovery. The document clustering is required for this work. It can use self-similarity function between pair of document graph that similarity can use for clustering with the help of affinity propagation and efficiency of algorithm can be measure by F-measure function.