Error-tolerant similarity search at the Federal Ministry for Migration and Refugees (BAMF)

Solution

To reliably and quickly find data records on a person that were beyond biometric procedures, master data was made usable for similarity searches via the matchmaker technology platform. This facilitates case processing and reduces risks arising from the failure to find heterogeneous personal data.

Find similar person records

To get started with case processing in MARiS, matchmaker provides users with a single form field to enter search terms in an unstructured manner and find the relevant person records. The approximate search takes into account erroneous entries and different qualities in both the input and the reference data – regardless of script, language, format or transcriptions.

In addition to common phonetic transformations, matchmaker offers solutions for transliteration from writing systems such as Dewanagari, Hangul, Katakana and Mandarin. The specially developed canonicalization function for Arabic names abstracts a wide range of transcript variants. For example, the Latin letters (sequences) 'o', 'u', 'w', 'oo' and 'ou' corresponding to the same Arabic character 'و' are treated equivalently. If „محمد“ is usually transcribed as Muhammad in the Anglo-American language area, matchmaker also evaluates the variants Mohammed, Mohamed, Muhamed, Muhammed, Muhamet or Mohammad as a match.

The foundation for duplicate cleansing

Using the same algorithms that ensure data availability, clusters of potential person duplicates can be generated. Additional information, e.g. on family associations, is taken into account and visualized. In this way, duplicates can be detected automatically during data creation and preventive measures can be taken.