Conflicts Resolving for Fusion of Multi-source Data

Data exits in a variety of sources. To make use of data, we need to get a unified form of data from multi-source, this is what data fusion should resolve. However, conflicts may arise among these sources. Existing conflict resolution work does have solved some issues, but it's not enough. In this paper, we propose a more comprehensive truth discovery solution of conflicts from multiple sources. Our main job is to make out the source reliability, with consideration of confidence interval of the estimation. Meanwhile, correlation among entities is added since entities are not all independent, they are correlated more or less. In addition, the newer information counts more generally, so we give the newer entities a bigger factor in correlation computing. Experiments on real-world data sets show the advantages of the proposed method, which has more accurate values.