Truth Discovery on Categorical Multi-Valued Attributes with Source Relations

In the era of big data, a mass of data fusion applications need to obtain accurate and complete attribute information about various objects. Most existing truth discovery methods are for single-valued attributes, which will be invalid for multi-valued attributes. They are usually based on the assumption that the data sources are independent of each other. It is not in line with the reality that related relationship occurs between them. Since multi-valued attributes are at large discrete, this paper proposes a truth discovery method incorporating the qulity and the correlations of data sources, namely Categorical Multi-Truths Finder algorithm (CMTF), which iteratively derives the accuracy and the recall rate of data source along with the truth set for each attribute. The experimental results show that CMTF outperforms the state-of-the-art.