Acquiring knowledge from inconsistent data sources through weighting

This paper presents a formal framework for multiple data source (MDS) discovery. A measure is first proposed for estimating the consistency, inconsistency and uncertainty between data sources using possibilistic minimal model. Then, two metrics are defined for measuring the support and confidence of a set of formulae (itemsets) in terms of the degree of consistency of the items. The consistency measure, in conjunction with support-confidence framework in data mining, assists in identifying interesting knowledge from MDSs. Finally, the impact of consistency among knowledge bases is considered to determine the knowledge base from which a set of formulae is most likely identified as a pattern of interest. A major advantage of this framework is that the mining algorithm supports the reasoning about the knowledge from possibilistic data-sources. We evaluate the proposed approach with both examples and experiment, and demonstrate that our method is useful and efficient in identifying interesting patterns from multiple databases.