Improving Data Provenance Reconstruction via a Multi-Level Funneling Approach

The ease with which data can be created, copied, modified, and deleted over the Internet has made it increasingly difficult to determine the source of web data. Data provenance, which provides information about the origin and lineage of a dataset, assists in determining its genuineness and trustworthiness. Several data provenance techniques record provenance when the data is created or modified. However, many existing datasets have no recorded provenance. Provenance Reconstruction techniques attempt to generate an approximate provenance in these datasets. Current reconstruction techniques require timing metadata to reconstruct provenance. In this paper, we improve our multi-funneling technique, which combines existing techniques, including topic modeling, longest common subsequence, and genetic algorithm to achieve higher accuracy in reconstructing provenance without requiring timing metadata. In addition, we introduce novel funnels that are customized to the provided datasets, which further boosts precision and recall rates. We evaluated our approach with various experiments and compare the results of our approach with existing techniques. Finally, we present lessons learned, including the applicability of our approach to other datasets.