Irregular cellular learning automata-based algorithm for sampling social networks

Since online social networks usually have quite huge size and limited access, smaller subgraphs of them are often produced and analysed as the representative samples of original graphs. Sampling algorithms proposed so far are categorized into three main classes: node sampling, edge sampling, and topology-based sampling. Classic node sampling algorithm, despite its simplicity, performs surprisingly well in many situations. But the problem with node sampling is that the connectivity in sampled subgraph is less likely to be preserved. This paper proposes a topology–based node sampling algorithm using irregular cellular learning automata (ICLA), called ICLA-NS. In this algorithm, at first an initial sample subgraph of the input graph is generated using the node sampling method and then an ICLA isomorphic to the input graph is utilized to improve the sample in such a way that the connectivity of the sample is ensured and at the same time the high degree nodes are also included in the sample. Experimental results on real–world social networks indicate that the proposed sampling algorithm ICLA-NS preserves more accurately the underlying properties of the original graph compared to existing sampling methods in terms of Kolmogorov-Smirnov (KS) test.