Enhancing Social Media User Privacy
Open Access DepositedApplying Machine Learning-Based Anonymization and Pseudonymization Techniques
(i) a 302 k-record retail-transactions corpus combining structured and free-text fields, and (ii) a 1 k-record ad-click dataset capturing user-engagement signals. After a 70/30 train–test split and domain-specific preprocessing, the SpaCy model achieved 0.96 precision, 0.94 recall, and a 0.95 F1 score in detecting personally identifiable information (PII), outperforming rule-based baselines. Tokenization plus SHA-256 hashing lowered empirical re-identification risk by ≈ 90 % (retail) and 88 % (ad-click) while retaining ≥ 94 % of statistical fidelity. Subsequent k- anonymity (k = 5) with hierarchical generalization and selective suppression drove total risk reduction to 90 % and 88 %, respectively, with Aggregate Data Accuracy (ADA) of 95 % and 94 %. Predictive performance dropped by only 4 %–7 % (AUC/F1), confirming that strong privacy need not cripple marketing models. The work contributes
The accelerating collection and monetization of personal data on social-media and e- commerce platforms has intensified the tension between user privacy and the business value of fine-grained analytics. This study proposes and empirically validates a layered, machine-learning-driven framework that integrates high-recall Named Entity Recognition (SpaCy NER), cryptographically salted pseudonymization, and context-aware k- anonymity to protect sensitive information while preserving analytical utility for targeted advertising. Two representative datasets were employed
and (3) an interpretable dual-metric dashboard (Aggregate Risk Reduction vs. Aggregate Data Accuracy) that bridges legal requirements and practitioner VII needs. The findings demonstrate that combining modern NLP with classical anonymization delivers regulator-ready protection, mitigates user-privacy fatigue, and sustains the commercial viability of data-driven personalization.
(1) a reproducible privacy–utility frontier across heterogeneous data scales
(2) a modular architecture that can be extended with differential privacy or post-quantum hashes
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.