Multi-objective community detection method by integrating users' behavior attributes

Social networks usually have abundant attributes associated with users to describe their features. Behavior attribute is one of the most important types of attribute which can better reflect users' intrinsic interests. In practice, many network applications prefer communities that not only are densely intra-connected, but also have homogeneous attribute value on specific behavior attributes. Structure clustering and attribute categorization are two types of method which can take full advantage of structure information and attribute information to partition the network, respectively. In this paper, we propose a novel community detection method by realizing structure clustering technology and attribute categorization technology simultaneously. Specifically, structure clustering is realized by optimizing modularity which captures densely intra-connected nature of communities. As for attribute categorization, a new metric named as homogeneity is defined to achieve the goal that nodes within each community have homogeneous attribute value, while in different communities have diverse attribute values. A multi-objective optimization evolutionary mechanism is adopted to optimize modularity and homogeneity simultaneously. Extensive experiments on several real-world networks demonstrate that our method can get a set of community structures corresponding to different trade-offs between structure clustering and attribute categorization.