Utilizing Stacking for Feature Reduction in Graph-Based Genealogical Entity Resolution - Stephen Ivie

Entity resolution is the process of aggregating information about an entity such that information related to the same real world entity is treated as a single entity. Our research compares two metric based machine learning to this problem on a large, post-blocking genealogy database. Our original method involved comparing each data element in the intersection of two records and resolving every comparison to continuous score. These scores were then stored in a sparse table which served as inputs to an artificial neural network. Our new approach is to treat each record as a graph with collections. Utilizing the hierarchal nature of the graph, more advanced comparators are built from basic comparator metrics through training artificial neural networks on a set of comparators and allowing the network's regression output to serve as input to another neural network which aggregates multiple comparators. This new approach resulted in higher precision, recall, and accuracy.