The Use of Reference Graphs in the Entity Resolution of Criminal Networks

Abstract

Entity resolution (ER) is the detection of duplicated records within a dataset representing the same real-world entity. The importance of ER is amplified within law enforcement as criminal data, or criminal networks, has inherent uncertainty and ER inaccuracy incurs a high cost. Commercial ER solutions focus on fast and scalable resolution of obvious pairs of entities, rather than the more complex non-obvious pairs which are so critical to law enforcement. Here we outline the use of proper names represented as reference graphs - generated from an algorithm that conducts name similarity, logic-based pruning, and classification using community detection and a proper name origin algorithm. The resultant classes are used at indexing and decision management stages within an ER model to support the detection of non-obvious duplicate entities. Utility is clearly demonstrated through the application of the approach on three real-world datasets of varying origin, size, topology, and heterogeneity.