collect annot wikipedia entiti web text sayali kulkarni amit singh ganesh ramakrishnan soumen chakrabarti step keyword base search entiti base search suitabl token span spot document identifi refer real world entiti entiti catalog system propos link spot web page entiti wikipedia larg base local compat text spot textual metadata entiti recent system exploit inter label depend limit wai propos gener collect disambigu approach premis coher document refer entiti relat topic domain give formul trade local spot entiti compat measur global coher entiti optim entiti assign np hard investig practic solut base local hill climb round integ linear program pre cluster entiti local optim cluster experi involv hundr manual annot web page ten thousand spot approach significantli outperform recent propos algorithm 