Learning Robust Web Wrappers

A main challenge in wrapping web data is to make wrappers robust w.r.t. variations in HTML sources, reducing human effort as much as possible. In this paper we develop a new approach to speed up the specification of robust wrappers, allowing the wrapper designer to not care about detailed definition of extraction rules. The key-idea is to enable a schema-based wrapping system to automatically generalize an original wrapper w.r.t. a set of example HTML documents. To accomplish this objective, we propose to exploit the notions of extraction rule and wrapper subsumption for computing a most general wrapper which still shares the extraction schema with the original wrapper, while maximizes the generalization of extraction rules w.r.t. the set of example documents.