Naive Bayes のバランス入力を使用した特許分類

抄録

In this preliminary study, we developed a Naive Bayes text classifier to predict a patent document's Cooperative Patent Classification (CPC) codes.　Initially we took each CPC code as a prediction class, however, classes with a small amount of training data were not predicted well.　Therefore, we developed a method to produce new sets of prediction classes that improve classifier performance by grouping neighboring CPC codes together into classes of balanced size.　Moreover, we discuss further work to create sets of prediction classes from clusters in citation networks.