POSITIVE SEMIDEFINITE SIMILARITY MEASURE ESTIMATION PROGRAM FOR MASS INSTANCES, RECORDING MEDIUM AND DEVICE

摘要

PPROBLEM TO BE SOLVED: To provide a program for estimating a similarity measure between arbitrary instances in a database consisting of a large number of instances, and to provide a recording medium and a device. PSOLUTION: This similarity measure estimation program is composed of steps of setting an instance set consisting of the large number of instances; setting a selected instance set by selecting one instance; calculating a positive semidefinite similarity measure matrix consisting of the similarity measure between the selected instance and all other instances; calculating correction Cholesky decomposition of the matrix; calculating characteristic values of all other instances by a decomposition matrix; determining the magnitude of a maximum value of the characteristic values and a threshold value of the characteristic values; adding the other instances to the selected instance set when the maximum value is large; terminating repetition when the maximum value is smaller than the threshold value by repeating each step; and estimating the similarity measure of two instances by the inner product of an instance vector concerning the two optional instances other than the selected instance set. The recording medium storing this program, and an estimating device are also provided. PCOPYRIGHT: (C)2008,JPO&INPIT