Truth Finding with Attribute Partitioning

Abstract

Truth finding is the problem of determining which of the statements made by contradictory sources is correct, in the absence of prior information on the trustworthiness of the sources. A number of approaches to truth finding have been proposed, from simple majority voting to elaborate iterative algorithms that estimate the quality of sources by corroborating their statements. In this paper, we consider the case where there is an inherent structure in the statements made by sources about real-world objects, that imply different quality levels of a given source on different groups of attributes of an object. We do not assume this structuring given, but instead find it automatically, by exploring and weighting the partitions of the sets of attributes of an object, and applying a reference truth finding algorithm on each subset of the optimal partition. Our experimental results on synthetic and real-world datasets show that we obtain better precision at truth finding than baselines in cases where data has an inherent structure.