Knowledge-Based Entity Resolution with Contextual Information Defined over a Monoid
Entity resolution (aka record linkage) addresses the problem to decide whether two entity representations in a database or stream correspond to the same real-world object. Knowledge-based entity resolution is grounded in knowledge patterns, which combine rules defined by Horn clauses with conditions prescribing when the rule is applicable, and conditions specifying when the application of the rule is not permitted. So far, these positive and negative conditions are expressed as bindings of the variables appearing in the Horn clause. In this paper the condition part of a knowledge pattern is generalised to a context, which is still defined by a positive and a negative part, but for both equations involving operators are permitted. The paper concentrates on conditions over a monoid for the constraints in a context. With this generalisation standard properties of knowledge patterns such as minimality, containment and optimality are investigated, which altogether minimise redundancy and thus optimise the inference of equivalences between entities.
- 1.Arasu, A., Götz, M., Kaushik, R.: On active learning of record matching packages. Proceed. SIGMOD 2010, 783–794 (2010)Google Scholar
- 5.Dong, X., Halevy, A., Madhavan, J.: Reference reconciliation in complex information spaces. Proceed. SIGMOD 2005, 85–96 (2005)Google Scholar
- 10.Sarawagi, S., Bhamidipaty, A.: Interactive deduplication using active learning. In: KDD, pp. 269–278 (2002)Google Scholar