Improving Social Network Analysis to Enhance the Identification of Influential Nodes

Abstract

In online social networks, social influence plays a vital role. Information in social networks propagates virally, as a consequence of this social networks are used to spread influence for multiple reasons including viral marketing, behavioral adoption, and opinion propagation. Numerous researchers are taking action to tackle this social influence study, including initial spreader detection, influence maximization, and influencer rankings, but there are numerous areas which are still quiet challenging. Detecting the influential nodes that occupy significant positions in social networks is a substantial problem as it relates to the effective distribution of information and has wide applications. Traditional ranking algorithms generally target only one out of global, local or community features. Global centrality is mostly measured in terms of betweenness centrality, which is deceptive as it assigns equal value to nodes of high degree scores which are central to local community and global bridges which connect different communities. Moreover, local centrality is usually measured by traditional degree centrality algorithm, which only considers the number of the nearest neighbors. We have used local centrality algorithm which take into consideration the number of the nearest and the next nearest neighbors of node. The thesis proposes a novel ranking framework in which we have taken into account both global and local features to measure influence. Global diversity is measured in terms of proposed bridge centrality method and local centrality is measured using local centrality method. The proposed approach is applied and tested on four different datasets. The thesis used a cross validation technique to measure the accuracy of proposed method. The hybrid classifier achieves 99% accuracy which is up to the mark. The overall aim of our research is to improve social network analysis to enhance the identification of influential nodes/ key players. The experimental results demonstrate that the proposed technique can rank influential nodes efficiently and accurately on social networks.