The Importance of Being Earnest: Social Sensing With Unknown Agent Quality
Stefano Maranò, Vincenzo Matta, Peter Willett · IEEE Transactions on Signal and Information Processing over Networks · 2016
We consider parametric estimation from social data, exploiting the generative social sensing model proposed in [18], [19]. First, we provide a detailed analysis of the estimation performance bounds, in terms of the Fisher information matrix, with emphasis on the regimes where the number of network agents and/or the number of monitored agents' activities is large. Then, we examine the performance of two likelihood-based estimation algorithms, namely: the expectation-maximization algorithm, and the Fisher scoring method, which both achieve the aforementioned estimation performance bounds. The analysis, corroborated by the application to a couple of classical estimation problems, allows: 1) highlighting the fundamental scaling laws for the considered parametric social sensing model; 2) identifying viable algorithmic procedures that can be useful even in large dataset applications.