Detecting Basic Level Categories by Term Weighting and Feature Entropy

Abstract

With the explosive growth of wide variety resources in the real world, data structure mining becomes a meaningful subject. In cognitive psychology, there is a family of categories called basic level categories. This method can reflect natural categories of corpus faithfully. These categories represent the most nature level; neither too general nor too specific. People frequently prefer to use basic level concepts in their daily life. Basic level concepts are the abstraction of basic level categories. According to the study of cognitive psychology, we find that basic level categories play an important role in structural hierarchy relationship for human to understand. Existing methods can find out basic level categories in corpus but cannot work in continuous datasets. This paper proposed a method which can improve the similarity representation of category utility and help finding basic level categories not only in text datasets but also in continuous datasets. Our experiments demonstrate that our method has good performance in both two kinds of datasets than mainstream model.