An Investigation of Linkage in Nominal Data

The Internet provides access to vast volumes of nominal data (data containing names) collected for a range of different purposes (e.g. parish registers containing baptism, marriage, and burial records). To mine these data effectively methods must exist, that are aware both of the source and semantics of the data, as well as the types of linkage (relationships between records) that can exist. Furthermore, as well as handling the implicit constraints of nominal data, such a system must also be able to handle automatically a range of temporal and spatial rules and constraints. This paper describes some initial research into this problem. It describes a prototype nominal data workbench that allows the specification and examination of several linkage types and discusses the merits of alternative name matching methods. The paper concludes that in the cases examined so far, effective nominal data linkage is essentially a query optimization process. The process is made more efficient if linkage specific indexes exist, and suggests that query re-organization based on these indexes, though a complex process, are entirely feasible. To facilitate the use of indexes and to guide the optimization process, the work suggests use of formal ontologies. Keyword: nominal data; record linkage; linkage workbench; domain ontologies; name matching; data mining 1. A Workbench Approach Access to vast volumes of nominal data [6] (data associated with names e.g. birth/death records, parish records, census data, text articles, newspapers, multimedia, etc) is now readily available through the Internet. Mining these data resources involves linkage [33] (how two names are related, e.g. similar surname, same spatio-temporal location, legal association, etc). In this paper are presented the results of the first year of a three-year Ph.D. study into automating various aspects of nominal data linkage. In particular this paper is concerned with algorithms associated with matching one of the key elements of nominal data linkage. The human approach to nominal data linkage is to acquire a corpus of knowledge, which is then used to compare entire records, and to rank matches based on experience. Current computer based techniques use a more or less complicated matching algorithm to search a database (or databases) for linkage and then to rank the linkage probabilities. All prior knowledge is captured in the program logic, and the program is usually optimized for speed or accuracy. Current systems are unable: • dynamically to modify their prior knowledge, • arbitrarily to chose which datasets to search, • to vary the matching algorithms used and dynamically to change their control parameters, • to combine the results from disparate nominal data linkage / matching run. What is required is a workbench approach which has components that may be tailored for different situations and parameterized by the user to meet changing requirements. The goal of this research is to investigate the elements of such a workbench and to evaluate prototypes using the extensive Lancashire and Cheshire Parish Register [14] archive held on the MIMAS database computer located at Manchester University. Capturing and processing prior knowledge in the workbench approach is dependent on combining