Fine-Grained Personalized Data Aggregation Scheme with High Quality and Privacy Protection.

Area of Science:

• Computer Science
• Ubiquitous Computing
• Data Privacy

Background:

• Mobile crowd sensing (MCS) requires truth discovery for aggregating noisy data.
• Existing methods often compromise user privacy or lack task-specific reliability.

Purpose of the Study:

• To develop a privacy-preserving truth discovery framework for MCS.
• To enable high-quality data aggregation while protecting user location and data privacy.
• To support fine-grained, task-level incentives.

Main Methods:

• A task-wise, personalized, privacy-preserving truth discovery framework.
• Per-user, per-task weight learning for aggregation.
• Paillier homomorphic encryption for aggregate-only processing.
• Task-scoped unlinkable pseudonyms for structural privacy.

Main Results:

• Achieved high accuracy (MAE/RMSE ~10-5) compared to non-private baselines.
• Demonstrated fast and stable convergence.
• Showed predictable scaling with users, tasks, and key sizes.
• Identified cloud-side decryption as the main computational cost.

Conclusions:

• Personalized weighting and structural privacy offer practical, high-quality aggregation for privacy-critical MCS.
• The framework effectively balances data utility with robust privacy protection.
• Enables reliable data aggregation in MCS without compromising individual privacy.