Title:Partitioned Sampling of Public Opinions Based on Their Social Evolution

Abstract:Public opinion polling is typically done by random sampling from the entire population, treating the opinions of individuals as independent. In the real world, individuals' opinions are often correlated, especially among friends in a social network, due to the effect of both homophily and social influence. In this paper, we propose a partitioned sampling method, utilizing the correlations between individuals' opinions to improve the sampling quality. In particular, we propose an adaptation of an opinion evolution model in social networks, and formulate an optimization problem based on this model as finding the optimal partition for the partitioned sampling method to minimize the expected sample variance of the estimated result. For the opinion evolution model, we develop an efficient and exact computation of opinion correlations between every pair of nodes in the social network. For the optimization task, we show that when the population size is large enough, the complete partition which contains only one sample in each component is always better, and utilize the correlation computation result obtained earlier to reduce finding optimal complete partition to a well-studied Max-r-Cut problem. We adopt the semidefinite programming algorithm for Max-r-Cut to solve our optimization problem, and further develop a greedy heuristic algorithm to improve the efficiency. We use both synthetic and real-world datasets to demonstrate that our partitioned sampling method results in significant improvement in sampling quality.