SDDup: Confidentiality-Aware Semantic Deduplication

Full-document encryption in existing document management systems creates major storage bottlenecks and hampers fine-grained access control. To overcome this, we present SDDup, a novel model that employs semantic-aware segmentation to separate common from unique, sensitive data segments. This allows for efficient deduplication and targeted encryption while maintaining integrity and regulatory compliance. We validate SDDup through theoretical and empirical analysis, security evaluation, and large-scale experiments on Brazilian birth certificates and university degrees. We compare SDDup against EDRStore, a leading system for generic encrypted data reduction. Our findings highlight SDDup as a competitive and scalable solution for document management.