Entity correspondence with second-order Markov logic

Abstract

Entity Correspondence seeks to find instances that refer to the same real world entity. Usually, a fixed set of properties exists, for each of which the similarity score is computed to support entity correspondence. However, in a knowledge base that has properties incrementally recognized, we can no longer rely only on the belief that two instances sharing value for the same property are likely to correspond with each other: a pair of different properties that are of hierarchies or specific relations can also be evidential to corresponding instances. This paper proposes the use of second-order Markov Logic to perform entity correspondence. With second-order Markov Logic, we regard properties as variables, explicitly define and exploit relations between properties and enable interaction between entity correspondence and property relation discovery. We also prove that second-order Markov Logic can be rephrased to first-order in practice. Experiments on a real world knowledge base show promising entity correspondence results, particularly in recall. © 2013 Springer-Verlag.