Towards Real-Time Anonymization of Unstructured Data with Deep Learning: A Privacy-Preserving Framework Prototype
Year: 2026
Type: conference-paper
Abstract: The widespread deployment of data systems across cloud-based and distributed digital environments has exposed fundamental limitations in existing anonymization solutions, which are typically implemented as isolated preprocessing mechanisms lacking semantic reasoning, formal policy enforcement, and cryptographic guarantees. This paper proposes a Privacyby-Design framework for the real-time anonymization of unstructured data, integrating deep learning, ontology-based compliance modelling, fine-grained authorization, and attributebased encryption within a unified architectural approach. The framework employs deep learning techniques for the contextual detection of sensitive entities, which are semantically enriched through alignment with the Data Privacy Vocabulary (DPV), enabling machine-interpretable regulatory reasoning. Privacy decisions are dynamically enforced via an Attribute-Based Access Control (ABAC) layer, while Ciphertext-Policy Attribute-Based Encryption (CP-ABE) binds access policies directly to encrypted data, ensuring end-to-end confidentiality beyond conventional redaction techniques. In contrast to existing approaches that treat anonymization, access governance, and encryption as disjoint processes, the proposed solution adopts a formally structured, modular microservice architecture designed for low-latency inference, while preserving semantic integrity and regulatory compliance. The main contribution lies in the integration of AIdriven entity recognition, ontology-grounded policy reasoning, and cryptographic enforcement, thereby establishing a scalable and interoperable foundation for the secure, real-time processing of unstructured data in regulated digital domains. A preliminary benchmark with 11,142 prototype executions across six NER/ML models provides initial empirical evidence of feasibility, traceability, and runtime behavior.
Institution: National Polytechnic School
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7214734700