An Improved Chinese String Comparator for Bloom Filter Based Privacy-Preserving Record Linkage.

Area of Science:

• Computer Science
• Information Security
• Data Linkage

Background:

• Data sharing is crucial but raises privacy concerns.
• Existing privacy-preserving record linkage (PPRL) methods are primarily designed for alphabetic languages.
• Chinese language presents unique challenges for PPRL due to its character-based nature.

Purpose of the Study:

• To develop and evaluate a novel PPRL technique tailored for the Chinese language environment.
• To address the limitations of existing PPRL methods in non-alphabetic contexts.
• To enhance data linkage accuracy while ensuring privacy.

Main Methods:

• Chinese characters are encoded into SoundShape codes based on pronunciation and shape.
• SoundShape codes are encrypted using a Bloom filter.
• Record similarity is calculated using Dice similarity on encrypted fields.
• The impact of Bloom filter's false positive rate and sound/shape code proportions were analyzed.

Main Results:

• The proposed method demonstrated improved precision, recall, and F1-score on synthetic datasets.
• Performance was evaluated across various false positive rates and sound/shape code proportions.
• The approach achieved high-quality classification results for Chinese PPRL.
• The method offered a relatively low additional computational cost.

Conclusions:

• The developed SoundShape code and Bloom filter-based PPRL method is effective for Chinese language data.
• This technique successfully balances data linkage accuracy with privacy preservation.
• The findings suggest a viable solution for PPRL in non-alphabetic language environments.