Tracer : a machine learning approach to data lineage

Abstract

The data lineage problem entails inferring the source of a data item. Unfortunately, most of the existing work in this area relies either on metadata, code analysis or data annotations. In contrast, our primary focus is to present a machine learning solution that uses the data itself to infer the lineage. This thesis will formally define the data lineage problem, specify the underlying assumptions under which we solved it, as well as provide a detailed description of how our system works.