Identifying influencers who can spread information to many other individuals from a social network is a fundamental research task in the network science research field. Several measures for identifying influencers have been proposed, and the effectiveness of these influence measures has been evaluated for the case where the complete social network structure is known. However, it is difficult in practice to obtain the complete structure of a social network because of missing data, false data, or node/link sampling from the social network. In this paper, we investigate the effects of node sampling from a social network on the effectiveness of influence measures at identifying influencers. Our experimental results show that the negative effect of biased sampling, such as sample edge count, on the identification of influencers is generally small. For social media networks, we can identify influencers whose influence is comparable with that of those identified from the complete social networks by sampling only 10-30% of the networks. Moreover, our results also suggest the possible benefit of network sampling in the identification of influencers. Our results show that, for some networks, nodes with higher influence can be discovered from sampled social networks than from complete social networks.

Several indices for estimating the influence of social media users have been proposed. Most such indices are obtained from the topological structure of a social network that represents relations among social media users. However, several errors are typically contained in such social network structures because of missing data, false data, or poor node/link sampling from the social network. In this paper, we investigate the effects of node sampling from a social network on the effectiveness of indices for estimating the influence of social media users. We compare the estimated influence of users, as obtained from a sampled social network, with their actual influence. Our experimental results show that using biased sampling methods, such as sample edge count, is a more effective approach than random sampling for estimating user influence, and that the use of random sampling to obtain the structure of a social network significantly affects the effectiveness of indices for estimating user influence, which may make indices useless.