Identification of Neutral Members in Social Networks using a Distance and Fuzzy based Model

Abstract

Understanding community structure helps to interpret the role of actors in a social network. Actor has close ties to actors within a community than actors outside of its community. Community structure reveals important information such as central members in communities and bridges members who connect communities. Clustering algorithms like hierarchical clustering, affinity propagation, modularity and spectral graph clustering had been applied in social network clustering to identify community structures in it. This study proposes a novel method for distance measurement between nodes and centroids. Distance is measured based on the shortest path length and number of common nearest neighbors with one path length. This measure, "Proportional closeness" is used to assign nodes to the closest centroid. A fuzzy system is also applied to find the closest centroid to a node when similar proportional closeness values are present for multiple centroids. The method has been applied to two artificial networks and one real world network data to test its accuracy on membership identification. The results revealed that the method successfully assigns members to its nearest centroid and leave neutral members aside without assigning to any centroid.