OPTIMAL CHARACTERISTIC SELECTION METHOD FOR ENABLING MULTI-LABEL PATTERN CLASSIFICATION, DEVICE THEREOF, AND MULTI-CATEGORY DOCUMENT CLASSIFICATION DEVICE

摘要

Disclosed is an optimal characteristic selection method for enabling multi-label pattern classification. According to an embodiment of the present invention, the optimal characteristic selection method for multi-label pattern classification includes the steps of: generating a characteristic subset including first characteristics selected from an entire set of characteristics including all the multiple characteristics as configuration elements to be used for the label classification of multiple patterns individually; calculating a characteristic evaluation value regarding each of the multiple second characteristics not included in the characteristic subset in the entire set of the characteristics based on a first correlation between a first label and the second characteristics and a second correlation between a second label and the first and second characteristics; and selecting one of the second characteristics as an optimal characteristic to be added to the characteristic subset based on the characteristic evaluation value.;COPYRIGHT KIPO 2017