joint optim wrapper gener templat detect shuyi zheng ruihua song ji rong wen di wu websit larg collect page gener dynam underli structur sourc databas data categori typic encod similar page common script templat recent year ad servic comparison shop vertic search specif domain motiv research extract technolog high accuraci previou work assum input page wrapper induct system conform common templat easili identifi term common schema url observ hard distinguish templat dynam url todai extract accuraci heavili depend consist input page argu riski determin page share common templat sole base url propos approach util similar page detect templat approach separ page notabl differ gener wrapper experiment result show propos approach feasibl effect improv extract accuraci 