Towards reliable social sensing in cyber-physical-social systems

Abstract

Social sensing is a new application paradigm of cyber-physical-social systems (CPSS), where a group of individualsvolunteer to report their claims about the physicalenvironment using cyber devices. A fundamental problem insocial sensing application is to ascertain source reliability andthe claim correctness without knowing either of them a priori, which is referred to as truth finding. Several key challengesexist in order to solve the truth finding problem. First, neitherthe source reliability nor the correctness of collected data areknown a priori. Second, data sources may make their claimswith different degrees of uncertainty and confidence. Third, it is challenging to accurately quantify the quality of truthfinding results without knowing the ground truth information. In this paper, we develop a confidence-aware truth findingscheme to address the above challenges under a unified analyticalframework. The confidence-aware truth finding scheme solvesa constraint estimation problem to jointly estimate both thesource reliability and claim correctness by explicitly consideringsource confidence and uncertainty on the reported claims. Torigorously quantify the accuracy of the MLE estimation, wealso derive confidence bounds on the estimation results. Finally, we evaluate our confidence-aware scheme with techniques fromcurrent literature through an extensive simulation study. Theevaluation results validates the performance gains achieved byour proposed solution compared to other baselines.