Privately vertically mining of sequential patterns based on differential privacy with high efficiency and utility.

Area of Science:

• Data Mining and Privacy-Preserving Technologies
• Computer Science
• Information Security

Background:

• Sequential pattern mining is crucial for data analysis but raises privacy concerns due to sensitive sequential data.
• Existing differential privacy (DP) solutions often use inefficient horizontal mining algorithms, limiting accuracy and efficiency.
• There is a need for privacy-preserving sequential data analysis that balances strong privacy guarantees with high performance.

Purpose of the Study:

• To propose a novel private sequential pattern mining scheme, privVertical, that combines vertical mining with differential privacy.
• To enhance the efficiency and accuracy of privacy-preserving sequential pattern mining.
• To ensure the proposed method satisfies rigorous privacy standards.

Main Methods:

• Developed privVertical, a scheme integrating vertical mining algorithms with differential privacy.
• Introduced a differentially private HashMap (privHashMap) using the Sparse Vector Technique to record noisy support counts.
• Implemented pre-pruning of infrequent candidate sequences to reduce noise and improve accuracy.

Main Results:

• PrivVertical avoids costly database scans and projections inherent in horizontal methods.
• The privHashMap and Sparse Vector Technique improve the accuracy of frequent item recording.
• Experiments demonstrate that privVertical achieves higher accuracy and efficiency at equivalent privacy levels.
• Theoretical analysis confirms privVertical satisfies ε-differential privacy.

Conclusions:

• PrivVertical offers an efficient and accurate solution for privacy-preserving sequential pattern mining.
• The proposed method effectively addresses the limitations of existing DP solutions based on horizontal mining.
• This approach enables sensitive sequential data analysis with strong privacy guarantees.