SkipLPA: An Efficient Label Propagation Algorithm for Community Detection in Sparse Network

Abstract

The propagation phase of label propagation algorithm is a computationally intensive process and overall performance of algorithm depends on it. This phase determines the label of all the nodes by processing nodes recursively in the network. Rather processing all the nodes if it is possible to skip certain nodes from the propagation phase, then the process will speed-up. We propose an efficient algorithm SkipLPA based on label propagation algorithm for the discovering community structure in the sparse network. The initialization phase is split into two sub-phases. First sub-phase: only certain nodes are initialized with unique labels. Second sub-phase: remaining nodes will get initial labels from connected nodes and excluded from the propagation phase. The algorithm is tested not only on benchmark networks but also on the real world networks, and efficiently recovers community structure. The performance of this algorithm improves drastically without compromising the quality of community detected, as well as the number of iterations are reduced by skipping certain nodes from the propagation phase.