A framework for specifying explicit bias for revision of approximate information extraction rules