Using Data Across Dissimilar Domains with Technical and Legal Assurance of Privacy

As online activities expand across diverse domains, generating extensive datasets, the need for data grows-not only across a few similar domains but also across many dissimilar domains. However, this need faces two problems: data in dissimilar domains describe neither the same people nor the same features, and cross-domain transfer of data containing personal information requires privacy protection. We propose combining data in different domains through common latent clusters instead of common people or features. Privacy is assured both technically and legally by transforming data into statistics, applying fully homomorphic encryption to the statistics, computing cross-domain co-clustering with the encrypted statistics, and applying differential privacy. We demonstrate how our proposed method can use data across dissimilar domains while assuring technical and legal privacy.