hierarch topic segment websit ravi kumar kunal punera andrew tomkin paper problem identifi segment topic cohes region url tree larg websit page websit assum topic label distribut topic label gener standard classifi develop set cost measur character benefit accru introduc segment site base topic label propos gener framework measur describ qualiti segment provid effici algorithm find segment framework extens experi human label data confirm sound framework suggest judici choic cost measur algorithm perform surprisingli accur topic segment 