SyntheRela: A Benchmark For Synthetic Relational Database Generation

TMLR Paper6252 Authors

Abstract: Synthesizing relational databases has started to receive more attention from researchers, practitioners, and industry. The task is more difficult than synthesizing a single table due to the added complexity of relationships between tables. For the same reasons, benchmarking methods for synthesizing relational databases introduces new challenges. Our work is motivated by a lack of an empirical evaluation of state-of-the-art methods and by gaps in the understanding of how such an evaluation should be done. We review related work on relational database synthesis, common benchmarking datasets, and approaches to measuring the fidelity and utility of synthetic data. We combine best practices, a novel robust detection metric, and a novel approach to evaluating utility with graph neural networks into a benchmarking tool. We use this benchmark to compare 6 open-source methods over 8 real-world databases, with a total of 39 tables. The open-source \syntherela benchmark is available on GitHub with a public leaderboard.

Submission Type: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Long_Chen8

Submission Number: 6252