Influence percolation method for overlapping community detection

Communities in social networks represent social circles, and people within same circles often highly interact and strongly influence one another, and hence individual behaviors percolate quickly, and tend to invoke a resonance phenomenon, i.e., collective behaviors. Nowadays, boundaries between circles are more and more indistinct because people probably involve more than one circle. This paper develops an influence percolation method (IPM) for identifying overlapping communities. In IPM, we first determine the influenced area of each node through many times of simulations for influence percolation so that activated nodes with a frequency belong to the area, and those as clusters can initialize a cover for a network. Then, the cover is further refined through three stages, i.e., filtration, absorbtion and selection to determine communities. We systematically evaluate our method on plenty of artificial networks with various network characteristics as well as real-world networks. The results indicate that our method achieves the best performance on the networks with stronger overlaps, e.g., up to 50% overlapping nodes, each of which belongs to more than four communities, compared with the state of the art algorithms. An interesting finding is that two nodes tend to be indivisible if one is a seed, influence percolates into the other exceeding a certain frequency, and this threshold is mainly determined by the networks’ density.