Sifting Truths from Multiple Low-Quality Data Sources

In this paper, we study the problem of assessing the quality of co-reference tuples extracted from multiple low-quality data sources and finding true values from them. It is a critical part of an effective data integration solution. In order to solve this problem, we first propose a model to specify the tuple quality. Then we present a framework to infer the tuple quality based on the concept of quality predicates. In particular, we propose an algorithm underlying the framework to find true values for each attribute. Last, we have conducted extensive experiments on real-life data to verify the effectiveness and efficiency of our methods.