A scanning method for detecting clustering pattern of both attribute and structure in social networks

• Published in: Physica A Vol. 445; pp. 295 - 309
• Main Authors: Wang, Tai-Chi, Phoa, Frederick Kin Hing
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2016
• Subjects: Community/cluster detection; Scan statistic; Scanning window; Social networks; Statistical significance; Structure and attribute cluster; Statistical significance; Social networks; Scanning window; Scan statistic; Community/cluster detection; Structure and attribute cluster
• ISSN: 0378-4371; 1873-2119
• DOI: 10.1016/j.physa.2015.10.009

Community/cluster is one of the most important features in social networks. Many cluster detection methods were proposed to identify such an important pattern, but few were able to identify the statistical significance of the clusters by considering the likelihood of network structure and its attributes. Based on the definition of clustering, we propose a scanning method, originated from analyzing spatial data, for identifying clusters in social networks. Since the properties of network data are more complicated than those of spatial data, we verify our method’s feasibility via simulation studies. The results show that the detection powers are affected by cluster sizes and connection probabilities. According to our simulation results, the detection accuracy of structure clusters and both structure and attribute clusters detected by our proposed method is better than that of other methods in most of our simulation cases. In addition, we apply our proposed method to some empirical data to identify statistically significant clusters. •We propose a new community detection method for network data via scan statistics.•Both the network structure and node attribute are taken into account in this method.•A frequentist likelihood ratio test is proposed for the significance of cluster existence.•This method outperforms the existing methods (like CESNA) in several common criteria.•We apply this method to three real-life demonstrated examples.