On sampling social networking services

This article aims at summarizing existing methods for sampling social networking services and proposing a faster confidence interval for related sampling methods. Social networking services (SNSs), such as Facebook and Twitter, are an important part of the current Internet culture. Collecting samples from these networks, therefore, is necessary for learning more about sociological or cultural issues. However, typical sampling methods for networks, such as nodebased or link-based methods, are not always feasible for social networking services. Alternate approaches such as snowball sampling or random walk (RW) are applied to gather information from social networking services more efficiently. Thus it is beneficial to compare various sampling approaches for SNSs under different circumstances. Making statistical inference from the gathered information constitutes another problem, including the determination of the sampling probabilities and the estimation of related attributes for sampled networks. Many valuable approaches have been proposed and refined, yet their numerical properties have not been studied sufficiently, some of which will be shown in the following. Simulations will be carried out in terms of respondent driven sampling (RDS), which has many variations and comprises several important network sampling methods. Although it was initially invented to detect hidden population in society, it has been generalized and widely discussed in the related literature. Based on these existing methods, the author also proposes some modification of existing estimation methods and the construction of a faster confidence interval which has not been covered in the current literature.