Technology News

Data Scientist Ke Zhang’s Research Explores Homomorphic Encryption for Privacy-Preserving Marketing

As the digital economy makes consumer data a central business asset, businesses face a growing question: how can they analyze customer data without unnecessarily exposing it? In the paper “Research on the Application of Homomorphic Encryption-Based Machine Learning Privacy Protection Technology in Precision Marketing” presented at the 2025 3rd International Conference on Data Science and Network Security (ICDSNS) and published in the IEEE proceedings, Ke Zhang examines homomorphic encryption as a way to support machine-learning analysis on encrypted consumer data. The work addresses a practical challenge in modern marketing: preserving the usefulness of consumer data while reducing privacy risk. As data breaches become more common and privacy regulations such as the European Union’s General Data Protection Regulation and the California Consumer Privacy Act shape corporate data practices, companies face pressure to balance analytical value with stronger data protection. Homomorphic encryption allows computation on encrypted data without decrypting it, but broader adoption has been limited by computational cost and the difficulty of handling real-valued data such as transaction amounts. Zhang’s research builds a privacy-centered customer-segmentation framework around the K-means clustering algorithm, run entirely in the encrypted, or ciphertext, domain. A rational-to-integer encoding method allows the system to compute on non-integer metrics while maintaining encryption efficiency, and a ciphertext-domain pipeline supports customer profiling end to end without exposing raw customer data. The paper presents a systematic integration of homomorphic encryption with marketing analytics models, including clustering-based segmentation and predictive modeling approaches. A key part of the study is its approach to cluster initialization. Adapting the K-means++ idea to encrypted data, the method selects initial cluster centers using squared Euclidean distances and a secure selection protocol designed to keep distance information hidden. Combined with fewer computation rounds and parallel processing, this approach aims to reduce the computational overhead that has often limited encrypted clustering in larger-scale settings. On public consumer datasets, including the UCI Bank Marketing data, the optimized method reported a precision of 0.92, a recall of 0.90, and an AUC of 0.94, close to an unencrypted baseline of 0.93, 0.91, and 0.95, while keeping data encrypted throughout the process. The study found that clustering was most effective when consumers were grouped into four segments. The paper also compares the framework with other privacy-preserving approaches, including federated learning and differential privacy. Federated learning can train models more quickly but may still expose information through model gradients, while differential privacy can reduce accuracy because of added noise. Zhang’s framework is presented as a practical approach for balancing privacy protection and analytical utility in commercial machine-learning applications. The study further outlines a potential advertising workflow in which merchants encrypt purchase histories, send the encrypted data to an outside computing environment for clustering, and decrypt the  resulting segmentation outputs only on their side. This workflow is designed with data-protection concerns in mind, including concerns reflected in privacy laws such as the California Consumer Privacy Act. Zhang is a data scientist whose work spans recommendation systems, precision marketing, explainable machine learning, and privacy-preserving technologies. Her related research includes a recommendation model combining knowledge graphs and LSTM networks to address data sparsity, as well as an interpretable cross-selling framework based on XGBoost and SHAP. In 2026, she also worked on integrating agent-based intelligence into an internal analytics dashboard for automated attribution and root-cause analysis.  By keeping consumer data encrypted while preserving much of the predictive performance used in marketing analytics, Zhang's study offers a practical example of how privacy-preserving machine learning can be adapted for commercial settings. For industries that depend on personalization while facing growing privacy expectations, the research points toward AI systems that can support both data utility and stronger consumer-data protection.