Computing That Serves

Recommender Systems for Family History Source Discovery

Derrick Brinton
MS Thesis Defense
Friday, November 3, 1:00 PM
3346 TMCB
Advisor: Christophe Giraud-Carrier

As interest in family history research increases, greater numbers of amateurs are participating in genealogy. However, finding sources that provide useful information on individuals in genealogical research is often a strenuous process, requiring expertise. Many tools have been employed to help genealogists better do their work, including many computer-based tools. Prior to this work, recommender systems have not yet been applied to genealogy, though their ability to navigate patterns in large amounts of data seems to be useful when employed in the genealogical domain.
We create the Family History Source Recommender System to mimic human behavior in locating sources of genealogical information. The recommender system is seeded with existing source data from the FamilySearch database. The typical recommender systems algorithms aren’t designed for family history work, so we adjust them to fit the problem. In particular, recommendations are created for deceased individuals, with multiple users being able to consume the same recommendations. Additionally, our similarity computation takes into account as much information about individuals as possible in order to create connections that would otherwise not exist. We use offline n-fold cross-validation to validate the results. The system provides results with relatively high accuracy.