Some methods for blindfolded record linkage（一些蒙上眼睛记录链接的方法）.pdfVIP

Some Methods for Blindfolded Record Linkage Author: Tim Churches and Peter Christen Presentation by Liyuan Han MathComputer Science Department, Emory Background The linkage of records which refer to the same entity in separate data collections is a common requirement in public health and biomedical research. If there is no shared common key, the partially-identifying attributes such as name, birth date, address are used. Someone need be give the access to match the records from different separate collections while keeping privacy protection Related work Anonymous record linkage/partially-identified record linkage : the specificity and overall efficiency of the linkage operation are diminished Blindfolded record linkage : the party undertaking the linkage is unable to see their actual values; good protection against a single party good protection against a single party Minimal-knowledge value comparisons Values are character strings and scalar quantities Comparison methods: match exactly, similarity comparison, bigram similarity cryptographically secure one-way hash functions N-gram similarity comparators The 2-gram in peter are pe, et, te and er ; in pete are pe, et, te, so the bigram score is 2 × 3/(4 + 3) = 0.86 Problem statement Alice, Bob: two separate data collections Carol: the third party trusted to determine match of A.a and B.a The actual value in A.a and B.a will not be revealed to any other parties Protocol 1 – step 1 Alice and Bob mutually agree on: Secret random key, shared only A and B Particular keyed hash transformation function Standard pre-processing for strings A mechanism for adding “chaff” to thwart statistic attack Protocol 1 – step 2 Pre-process the values by computing the set of bigrams Pro