automat record linkag seed nearest neighbour support vector machin classif peter christen task link databas import step increas number data mine project link data inform requir time consum expens collect specif data aim link match aggreg record refer entiti major challeng link larg databas effici accur classif record pair match match tradition classif base manual set threshold statist procedur recent develop classif method base supervis learn techniqu requir train data real world situat prepar manual expens cumbersom time consum process author previous present step approach automat record pair classif step approach train exampl high qualiti automat select compar record pair step train support vector machin svm classifi initi experi show feasibl approach achiev result outperform mean cluster paper variat approach present base nearest neighbour classifi improv svm classifi iter ad exampl train set experiment result show step approach achiev classif result unsupervis approach 