Regression Testing for Wrapper Maintenance

Recent workon Internet information integration ~sumesa library of wrappers, specialized informationextraction procedures. Maintainingwrappersis difficult, becausethe formatting regularities on whichthey rely often change. Thewrapperverification problemis to determine whethera wrapperis correct. Standard regression testing approachesare inappropriate, because both the formatting regularities anda site’s underlying content may change. Weintroduce RAPTURE, a fully-implemented, domain-independent verification algorithm. RAPTURE uses well-motivated heuristics to computethe similarity betweena wrapper’s expected and observed output. Experimentswith 27 actual Internet sites showa substantial performanceimprovementover standard regression testing.

Authors

• Nicholas Kushmerick Tessa Lau
• University College Dublin

Keywords

No keywords are indexed for this paper.