Improved Influence Factor Scheme for Detecting In- fluential Nodes in Mobile Phone Network

Abstract

The number of mobile phone users is increasing tremendously. Network of users are formed using the call (or social) interactions between these mobile phone users. Such networks could be represented using social network graphs where the nodes represent persons and the edges are the communications between them. In such networks, communi- ties of nodes with certain commonalities could be identified using community detection techniques. It should be noted that in every community there are usually nodes that have high influence, referred to as influential nodes. Knowledge of such nodes helps to under- stand the communities better and to relate with the community members. For example, re- moval of influential nodes from a criminal community will collapse the community and probably also the network they belong to. Also, influential nodes could be used to feed in- formation to an entire network. Therefore, it is important to accurately identify nodes that are prominent in a network. For these reasons, work on techniques for identifying influential nodes in communities is currently receiving attention in the research arena. One such tech- nique in literature is the influence factor scheme, which indicates how important an individ- ual node can be in a network. The scheme integrates betweenness centrality, closeness cen- trality and eigenvector centrality. However, the use of eigenvector centrality in the scheme strongly affects the measure of influence across the network by limiting the detection of influential nodes to the neighbouring nodes around the most influential nodes within the largest component (community) of the network. It neglects the fact that there could be an influential node in other smaller components in the network. This can be misleading, espe- cially in a massive social network like the mobile phone network that contains several neighbourhoods with hundreds or thousands of nodes and edges. This is because it is not necessarily true that every node that is connected to the most important nodes is truly im- portant. This limitation makes it difficult to detect the real influential nodes in large social networks. Principal component centrality is a variant of eigenvector centrality that considers every component (community) in a network when searching for influential nodes across a network graph. In this research, we present an improved influence factor scheme that incor- porates closeness centrality, betweenness centrality and principal component centrality to identify nodes that are truly influential in a mobile phone network. The improved scheme has better accuracy, precision and specificity. Furthermore, in terms of accessibility, the improved scheme outperforms the existing scheme because information through the detected influential nodes reached all members of the communities in the network.