Tackling the Redundancy and Sparsity in Crowd Sensing Applications

Driven by the proliferation of sensor-rich mobile devices, crowd sensing has emerged as a new paradigm of gathering information about the physical world. In crowd sensing applications, user observations are usually unevenly distributed across the monitored entities, and this gives rise to two major challenges -- redundancy and sparsity. On one hand, multiple users may observe the same entity, and their observations are sometimes conflicting with each other due to the unreliable nature of human-carried sensors. On the other hand, crowd sensing data are usually very sparse, and there may exist considerable number of entities that never receive any observations from users. Some existing work studies these two challenges separately. However, we can gain great benefits by dealing with them jointly. In this paper, we develop an integrated framework to estimate the true values of entities from redundant and sparse data in crowd sensing applications. In this framework, we propose an effective algorithm to infer the "missing" observations for each entity, and aggregate both user-contributed and inferred observations to discover the true values of entities. We conduct extensive experiments on real-world crowd sensing systems to demonstrate the advantages of the proposed framework on correctly inferring entity truths from redundant and sparse data.