Overlapping Community Detection Based on Node Correlation in Directed Complex Networks

With the rapid development of science and technology, there are all kinds of flow-based networks in our life, such as fraudulent fund transfer networks, etc. How to use the community detection algorithms to effectively detect the communities contained in various flow-based complex networks has become the key to the analysis of flow-based complex networks. There are a few algorithms for the community detection of the directed graph-based complex network. Most algorithms directly ignore the direction of edges or transform the directed network into the undirected network with weight when detecting communities in the directed complex network. Therefore, they lack effective ability to detect communities in flow-based fraudulent fund transfer networks. Therefore, in this paper, we propose a new concept of the correlation COR between nodes in flow-based fraudulent fund transfer networks, and based on this, a node COR-based local optimization algorithm (LCO) is designed. We adopt our algorithm for the problem of fraudulent funds-raising of the community detection of the victim in flow-based fraudulent fund transfer networks, in which a node represents a transfer record, which has four attributes: payment account, receiving account, transfer time and transfer amount, and node u points to node v with a directed edge, indicating that the receiving account of u is the same as payment account of node v. By analyzing the topological characteristics of each victim’s fund community, our algorithm chooses the node with the least entry degree as the seed node, and expands communities through the correlation between nodes. We verify the effectiveness of the LCO algorithm through the artificially generated directed networks. The results for the artificial directed networks show that our algorithm obtains higher NMI than undirected graph-based overlapping community detection algorithms.