This project is intended to put forward a new model and algorithm to deal with graph partitioning, which is an attractive part in the field of social network analysis. In the recent years, an exponent increasing number of studies have been undertaken to process social network data, partly as a result of the fact that so much social network data has become available. Another reason is that the significance has been public aware of in the sense of economic and research value. However, the results are not that perfect because the mathematical models before were simple. The real-world situation that one may belong to several groups should also be taken into consideration. In the mean time, overlapping communities will bring other problems for the graph partitioning algorithm. In this paper, we present a split-node method, a novel algorithm to detect overlapping communities in large data graphs, based on the summary of several classical graph partitioning algorithms.