Learning Approaches to Wrapper Induction

Thenumber,the size, andthe dynamicsof lntemetinformationsourcesbears abundantevidenceof the need of automationin informationextraction(IE). Thispaper deals withthe questionof howsuchextraction mechanismscanautomaticallybe createdbyinvokinglearning techniques. Theunderlying scenario of system-supportedIE is puttingcertainconstraintsonthe availabletrainingexamples.Therefore,the traditional approachesto formal languagelearningdo not capturethe kindof problems to be solvedwhenlearningthe correspondingextraction mechanisms. Weillustrate the resultingdifferencesbystudyingthe problemof learning a particular type of extraction mechanisms(so-called island wrappers).Weshowhow to decomposethis learningprobleminto different subproblemsthat canbe handledindependentlyandin parallel. Moreover,werelate the learning problemson handto the problemsthat learningtheorypapersoriginally addressandpoint out whatthey havein common andwherethe differencesare.

• Type: Conference
• Year: 2001
• Where: FLAIRS
• Authors: Gunter Grieser, Steffen Lange
• Keywords: Approachesto Formal Languagelearningdo, Artificial Intelligence, FLAIRS 2001, Mechanismscanautomaticallybe Createdbyinvokinglearning Techniques, Solvedwhenlearningthe Correspondingextraction Mechanisms