Robust Temporal Graph Clustering for Group Record Linkage

Research in the social sciences is increasingly based on large and complex data collections, where individual data sets from different domains need to be linked to allow advanced analytics. A popular type of data used in such a context are historical registries containing birth, death, and marriage certificates. Individually, such data sets however limit the types of studies that can be conducted. Specifically, it is impossible to track individuals, families, or households over time. Once such data sets are linked and family trees are available it is possible to, for example, investigate how education, health, mobility, and employment influence the lives of people over two or even more generations. The linkage of historical records is challenging because of data quality issues and because often there are no ground truth data available. Unsupervised techniques need to be employed, which generally are based on similarity graphs generated by comparing individual records. In this paper we present a novel temporal clustering approach aimed at linking records of the same group (such as all births by the same mother) where temporal constraints (such as intervals between births) need to be enforced. We combine a connected component approach with an iterative merging step which considers temporal constraints to obtain accurate clustering results. Experiments on a real Scottish data set show the superiority of our approach over a previous clustering approach for record linkage.