Privacy preserving record linkage meets record linkage using unencrypted data

IntroductionPrivacy preserving record linkage (PPRL) resolves privacy concerns because of its capabilities to link encrypted identifiers. It encrypts identifiers using bloom filters and performs record matching based on encrypted data using dice coefficient similarity. Matching data based on hashed identifiers impacts the performance of linkage due to loss of information.
 Objectives and ApproachWe propose a technique to optimize the bloom filter parameters and examine if the optimal parameters increase the performance of the linkage in terms of precision, recall, and f-measure. Let us consider a set of string values and calculate the similarity between any two of them using the Jaro-Winkler method. Now let us encrypt the string values using bloom filters and calculate the similarity between any two of them using the dice coefficient technique. Optimal parameters of bloom filters are those that minimize the difference between the calculated similarities using Jaro-Winkler vs. the calculated similarities using the dice coefficient technique.
 ResultsUsing publically available data, several first name and last name datasets each comprising 1000 unique values were generated.