T-PriDO: A Tree-based Privacy-Preserving and Contextual Collaborative Online Big Data Processing System

The emerging of big data era has urged the development of online big data processing systems. However, existing works seldom take the issue of privacy into account. In addition, the impacts of network structures among service providers are usually ignored. In this paper, we propose a cloud-based big data processing framework, where service providers are modeled as distributed cooperative learners predicting users’ preferences of items based on users’ contexts, while adapting the decision-making strategy based on users’ reward. We establish an item-cluster tree from top to the bottom to handle big data analysis. Service providers share information over a social network to complete collaborative learning. Considering the structure of the social networks among service providers, we also propose an adaptive algorithm to reduce the performance loss. Theoretical analysis shows that our proposal achieves sublinear regret and differential privacy of both network service providers and users. Experiments results validate that our proposed algorithms support increasing big datasets while strike a balance between privacy-preserving level and prediction accuracy.