Privacy-Preserving User Modeling for Digital Marketing Campaigns: The Case of a Data Monetization Platform

This work proposes using a collection of Deep Learning approaches to design the privacy-preserving data monetization platform to improve the digital marketplace where users have control over their data and marketers can identify the right users for their marketing campaigns through several steps. First, representation learning on hyperbolic space is performed to learn latent embeddings of user interests across multiple data sources with hierarchical structures. Second, Generative Adversarial Networks is performed to generate synthetic user interests from the embeddings. Third, the system adopts a Federated Learning technique to ensure the user modeling is trained in a decentralized manner on the user's own devices while keeping data localized without sharing with marketers. Last, a recommender system is built upon the learned user interests to identify the right users for digital marketing campaigns. Overall, this work provides a holistic solution for privacy-preserving user modeling for digital marketing campaigns.