Statistical Schema Learning with Occam's Razor
ACM SIGMOD International Conference on Management of Data, 2022
Abstract
We consider the problem of automatically creating a good schema for a denormalized table as an unsupervised machine learning problem. We define a principled schema optimization criterion based on Occam's razor that is robust to noise and readily extensible, and develop an efficient learning algorithm for it. Our approach runs 3 to 100 times faster than previous work while producing higher quality schemas, with roughly one fifth the errors.