| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide stable and high classification performance capable of following a change of appearance tendency of word at high speed. <P>SOLUTION: A document classification device divides an input document into sets of words, obtains a prior probability indicating importance of each class by an exponentially weighted moving average on the basis of a class belonging to the input document, stores word appearance information related to an appearance of each word included in the input document on the basis of the prior probability of each class, estimates a word long period appearance probability in each class of each word included in the input document, estimates a word short period appearance probability in each class of each word included in the input document by using the word appearance information, determines whether or not the word is a trend word on the basis of the word long period appearance probability and the word short period appearance probability of the word for each word included in the input document, and calculates a posterior probability which means the input document belongs to each class and classifies the input document into one or more classes on the basis of the determination result. <P>COPYRIGHT: (C)2013,JPO&INPIT |
