Police: An Effective Truth Discovery Method in Intelligent Crowd Sensing

摘要

With the progressively increasing number of smart mobile devices, mobile crowdsensing (MCS) becomes prevalent and pervasive in real life. People use devices as sensors to report claims about entities. Therefore, how to find the true information from the data uploaded by people is a key issue. Iterative updates, optimization or probabilistic models are three main aspects that most truth discovery focused. There is no denying the fact that these methods show their advantages and some limitations. They ignore the connection between the entities and focus on the data only in a single time node, without considering the trend of the data over a while. In this paper, we propose a new Probabilistic mOdel for real-vaLued sensing data on Correlated Entities named police. This model using time series analysis to predict the entity's probabilistic time distribution over a period of time. In this way, the efficiency of truth discovery can be improved. Moreover, this proposed model can be applied to correlated entities. If there are not have enough reliable users to observe entities, it is impossible to get accurate information, so we take the correlation among entities into consideration to ensure accuracy. Entities' association will increase the difficulty of solving the problem. However, we have proposed a timing grouping method, by dividing the entities into related groups and iterating through the block coordinate descent method. The experiments on real-world demonstrate that the proposed methods satisfy properties better than the existing truth discovery frame from conflicting information reported on correlated entities.