Abstract

In recent years, there has been a tremendous growth in the collection of digital information about individuals. Many organizations such as governmental agencies, hospitals, and financial companies collect and disseminate various person-specific data. Due to the rapid advance in the storing, processing, and networking capabilities of the computing devices, the collected data can now be easily analyzed to infer valuable information for research and business purposes. Data from different sources can be integrated and further analyzed to gain better insights. On one hand, the collected data offer tremendous opportunities for mining useful information. On the other hand, the mining process poses a threat to individual privacy since these data often contain sensitive information. In this thesis, we address the problem of developing anonymization algorithms to thwart potential privacy attacks in different real-life data sharing scenarios. In particular, we study two privacy models: LKC-privacy and differential privacy. For each of these models, we develop algorithms for anonymizing different types of data such as relational data, trajectory data, and heterogeneous data. We also develop algorithms for distributed data where multiple data publishers cooperate to integrate their private data without violating the given
privacy requirements. Experimental results on the real-life data demonstrate that the proposed anonymization algorithms can effectively retain the essential information for data analysis and are scalable for large data sets.