spam classification model generating apparatus and method, and program

摘要

PROBLEM TO BE SOLVED: To appropriately generate a spam determination model even as to practice data having a case example biased.;SOLUTION: A spam classification model generation device according to the present invention is configured to: calculate, when the case example having a spam label of any of a spam page (spam) and a non-spam page (ham) and a characteristic vector thereof is input, a loss of the case example in accordance with a size of a margin set for each class; store a set of a value of the loss and case example information in a spam label array or a ham label array in accordance with the spam label; eliminate a case example having a maximal loss from the label array; extract a characteristic vector of the case example having the maximal loss; update a weighted-vector by use of the characteristic vector; calculate a spam score by use of an updated weighted-vector and a characteristic vector stored in classification data storage means; determine the case example to be the spam when the spam score is a predetermined threshold value or more, and determine the case to be the ham when the spam score is less than the threshold value; and output a determination result.;COPYRIGHT: (C)2014,JPO&INPIT