Sampling Online Social Networks: Metropolis Hastings Random Walk and Random Walk

TL;DR

This paper introduces Metropolis-Hastings Random Walk and Random Walk with Jumps sampling methods for social network analysis, comparing their effectiveness in estimating various network properties and addressing data accessibility issues.

Contribution

The paper presents novel sampling strategies, MHRW and RWwJ, with theoretical foundations and empirical comparisons to improve social network data sampling.

Findings

MHRW outperforms RWwJ in estimating degree distribution and graph order.

RWwJ provides better estimates for follower/following ratios and mutual relationships.

MHRW reduces error rates in key network property estimations.

Abstract

As social network analysis (SNA) has drawn much attention in recent years, one bottleneck of SNA is these network data are too massive to handle. Furthermore, some network data are not accessible due to privacy problems. Therefore, we have to develop sampling methods to draw representative sample graphs from the population graph. In this paper, Metropolis-Hastings Random Walk (MHRW) and Random Walk with Jumps (RWwJ) sampling strategies are introduced, including the procedure of collecting nodes, the underlying mathematical theory, and corresponding estimators. We compared our methods and existing research outcomes and found that MHRW performs better when estimating degree distribution (61% less error than RWwJ) and graph order (0.69% less error than RWwJ), while RWwJ estimates follower and following ratio average and mutual relationship proportion in adjacent relationship with better results, with 13% less error and 6% less error than MHRW. We analyze the reasons for the outcomes and give possible future work directions.