Unveiling Dis-Integration

George Papadakis, Ekaterini Ioannou, Yannis Velegrakis

Abstract: Entity Resolution (ER) has been extensively studied over the last decade, with a plethora of algorithmic solutions, techniques, and methodologies having been proposed [1]. The individual state-of-the-art ER algorithms are offered through open-source systems, such as Magellan [2] and JedAI [3], which typically implement end-to-end solutions through a sequence of workflow steps. Each workflow step requires its own special configuration and fine tuning, thus turning the creation of complete ER solutions into a non-trivial, time-consuming process that requires adapting, among others, to the characteristics of the data to be resolved (e.g., relational, semi-structured, etc.), to its intrinsic noise (e.g., misspellings, abbreviations, etc.) as well as to application constraints (e.g., execution time).