Progressive Entity Resolution over Incremental Data

Entity Resolution (ER) algorithms identify entity profiles corresponding to the same real-world entity among one or multiple data sets. Modern challenges for ER are posed by volume, variety, and velocity that characterize Big Data. While progressive ER aims to efficiently solve the problem under time constraints by prioritizing useful work over superfluous work, incremental ER aims to incrementally produce results as new data increments come in. This paper presents algorithms that combine these two approaches in the context of streaming and heterogeneous data. The overall goal is to maximize the chances to spot duplicates to a given entity profile in a moment closest to its arrival time (early quality), without relying on any schema information, while being sufficiently efficient to process large volumes of fast streaming data without compromising the eventual quality (by cutting too many corners for efficiency). Experiments validate that our algorithms are the first to support incremental and progressive ER and, compared to state-of-the-art incremental approaches, improve early quality, eventual quality, and system efficiency by progressively and adaptively performing the unexecuted comparisons that are more likely to match when waiting for the next stream input increment.