Graph Sampling Approach for Reducing Computational Complexity of Large-Scale Social Network

TL;DR

This paper proposes a graph sampling method to reduce the size of large-scale social networks, aiming to decrease computational costs of network analysis while maintaining key properties.

Contribution

It introduces a sampling approach and compares different strategies like edge, node, and random walk sampling for large social networks.

Findings

Sampling strategies' effectiveness varies with network properties

Random walk sampling often preserves network metrics better

Sampling reduces computational complexity significantly

Abstract

Online social network services provide a platform for human social interactions. Nowadays, many kinds of online interactions generate large-scale social network data. Network analysis helps to mine knowledge and pattern from the relationship between actors inside the network. This approach is important to support predictions and the decision-making process in many real-world applications. The social network analysis methodology, which borrows approaches from graph theory provides several metrics that enabled us to measure specific properties of the networks. Some of the metrics calculations were built with no scalability in minds, thus it is computationally expensive. In this paper, we propose a graph sampling approach to reduce social network size, thus reducing computation operations. The performance comparison between natural graph sampling strategies using edge random sampling, node random sampling, and random walks are presented on each selected graph property. We found that the performance of graph sampling strategies depends on graph properties measured.