Truth discovery on crowd sensing of correlated entities

With the popular usage of mobile devices and smartphones, crowd sensing becomes pervasive in real life when human acts as sensors to report their observations about entities. For the same entity, users may report conflicting information, and thus it is important to identify the true information and the reliable users. This task, referred to as truth discovery, has recently attracted much attention. Existing work typically assumes independence among entities. However, correlations among entities are commonly observed in many applications. Such correlation information is crucial in the truth discovery task. When entities are not observed by enough reliable users, it is impossible to obtain true information. In such cases, it is important to propagate trustworthy information from correlated entities that have been observed by reliable users. We formulate the task of truth discovery on correlated entities as an optimization problem in which both truths and user reliability are modeled as variables. The correlation among entities adds to the difficulty of solving this problem. In light of the challenge, we propose both sequential and parallel solutions. In the sequential solution, we partition entities into disjoint independent sets and derive iterative approaches based on block coordinate descent. In the parallel solution, we adapt the solution to MapReduce programming model, which can be executed on Hadoop clusters. Experiments on real-world crowd sensing applications show the advantages of the proposed method on discovering truths from conflicting information reported on correlated entities.