Reputation-Based Truth Discovery With Long-Term Quality of Source in Internet of Things

Abstract

Although the Internet of Things (IoT) devices have been widely used for data collection in various applications, the observed data of an object from each IoT device (i.e., source) may vary from the ground truth due to the different qualities of IoT devices and sensing environments. Truth discovery has become a promising technology to extract the truth among multiple conflicting pieces of data from different sources. Existing methods usually assume the quality of source (source reliability) is unknown a priori and will be estimated as the weight for calculating the truth during each truth discovery task. However, in a long-term data observation scenario, the quality of source can be accumulated and utilized in the future truth discovery process. Aiming to take the long-term quality of source into consideration, in this article, we propose a reputation-based truth discovery method to derive the truth from the conflicting data. Specifically, we propose a generalized formulation with linear constraint for the truth discovery problem, which can cope with different regulations on the source reliability. Instead of directly using weight as the source reliability, we also define the reliability of a source by its contribution to the loss function. Then, we propose a novel reputation model to quantify the newly defined source reliability, which will be accumulated as the long-term source quality. Finally, we propose a reputation-based truth discovery model, where initial weights are assigned based on source reputations. Experiments conducted on real-weather conditions and GPS data sets demonstrate that our reputation-based truth discovery can reduce the number of iterations during truth discovery and achieve high accuracy.