Social Network Group Identification based on Local Attribute Community Detection

Abstract

Online social network has become an important platform for people's daily communication, information dissemination and sharing. The similarity of attribute characteristics such as content and behavior is very important to evolution and control of social network groups. Therefore, it is a challenge to consider the topology and attributes of network nodes in community identification effectively. In this paper, a social group identification method based on local attribute community detection (LA-CD) is proposed. a novel node pair similarity framework is proposed, and a novel local similarity distance factor is defined. It can eliminate the problem that the local similarity of nodes is too large due to the large number of adjacent nodes, and prevent the situation that having more neighbor sets can get lower local similarity values instead. Experiments on several real attributed ego-networks and artificial benchmark networks show that LA-CD can discover more real and effective network community than other state-of the-art approaches.