Parallel approach for network construction from large purchasing collections

Abstract

Community detection is one of the most relevant features of network-based models. Although community detection algorithms are capable of handling large datasets, this does not imply that there is no limit. When dealing with problems arising from applications such as customer purchasing interactions, building the network for a dataset comprised of millions of transactions will lead to some computational issues. In this paper, we tackle that computation challenge by using a parallel approach for network representations in presence of massive amount of purchasing data. The modularity measure is adopted to evaluate the convergence of community detection algorithms from the computed customer networks using multiple instance similarity functions and thresholding approaches. Numerical simulations using a real-world problem show the advantages of the proposed parallel solution.