| dc.description.abstract | Watermarking is used as proof of ownership for various data types such as images, videos, software, machine learning models, and databases. Datasets are crucial for data driven decision making using Machine Learning for tasks like prediction, recommendation, classification, and anomaly detection. Hence, it is not surprising that entire databases are being sold in data marketplaces. Protect- ing ownership rights upon such databases is, therefore, becoming increasingly important. Watermarking for relational databases has been an active field of research since 2002. However, how to water- mark non-relational databases involving complex data types has largely remained understudied. In this paper we revise previously proposed techniques for non-relational database watermarking and introduce an improved technique for graph database watermarking inspired by Zhuang et al. [28]. Our technique employs randomiza- tion to generate a watermark in an efficient manner that avoids the computational complex genetic algorithm optimization of Zhuang et al. We evaluated our technique in terms of performance, usability, security, and robustness by implementing it as a proof-of-concept. Our results showed that our technique is efficient, secure and robust against guessing and deletion attacks. | es |
