NANI: an efficient community detection algorithm based on nested aggregation of node influence

Abstract

Most traditional community detection algorithms were developed based on the similarity evaluation of topological features. None of them have concerned about the rules during the information dissemination from the dynamic aspect, and therefore, they cannot discover communities that best serve the purpose of information diffusion. In addition, since traditional algorithms have not considered the information interacting structure and mechanism with-between network nodes, the community center and backbone nodes identified may not even be effective in the actual information dissemination process. To deal with this problem, this paper explored the community detection issue from a different angle of dynamic information diffusion and did the following contributions: (1) proposing a novel model, called NI (Node Influence) to evaluate the influence between nodes and intermediate communities, (2) proposing a novel algorithm, NANI (Nested Aggregation of Node Influences) to merge the nodes or intermediate communities on the way upwards based on NI model, (3) conducting extensive experiments on all existing 9 kinds of similarity measures. Experiments showed that the NANI algorithm outperforms all the related methods at most cases, especially when the data volume scale is considerably huge.