On Multi-Network Sampling with Quality Guarantee

Hong-Han Shuai, De-Nian Yang, Chih-Ya Shen, Philip S. Yu, Ming-Syan Chen⋆arXiv 2013 · arXiv

Abstract

Online Social Networks (OSNs) have been gaining increasing importance in the past decade and inspired many researches for social network analysis (SNA). Currently, SNA algorithms are evaluated with real datasets obtained from large-scale real OSNs, which are usually sampled by Breadth-First-Search (BFS), Random Walk (RW), and some variations. However, BFS and RW have been proven to introduce bias, and thus it is more difficult for the analysis results on those datasets to provide useful insights for the OSNs. Most importantly, none of the released datasets have provided any statistical guarantees on the difference of the crawled datasets and the ground truth. On the other hand, all existing sampling algorithms only focus on crawling a single OSN, but each OSN is actually a sampling from the global offline social network. In other words, even if one crawls the whole dataset from a single OSN, the results may still be skewed and unable to fully reflect the properties of the global offline social network. In this paper, therefore, we make the first attempt and propose Quality-guaranteed Multi-network Sampler (QMSampler) to crawl and sample multiple OSNs jointly. QMSampler provides a statistical guarantee on the difference of the crawled real dataset and the ground truth (the perfect integration of all OSNs). Experimental results manifest that QMSampler generates much smaller bias than the existing approaches, and QMSampler is released as a free download.
BibTeX
@misc{shuai2013multi,
  author = {Hong-Han Shuai and De-Nian Yang and Chih-Ya Shen and Philip S. Yu and Ming-Syan Chen⋆},
  title = {{On Multi-Network Sampling with Quality Guarantee}},
  year = {2013},
  eprint = {1303.0157},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1303.0157},
}

Loading related content...

Figure