Keyword combination extraction in text categorization based on ant colony optimization

Due to the increasing number of documents in digital form, the automated text categorization (TC) has become more and more promising in the last ten years. A TC system can automatically assign a document with the most suitable category, but the reason for such an assignment is usually unknown by users. To make the TC system be interpretable, it is necessary to select a group of keywords, or termed a keyword combination, to describe each text category. In this paper, we propose a novel algorithm, keyword combination extraction based on ant colony optimization (KCEACO), to search the optimal keyword combination of a target category. By extending the traditional feature selection techniques, an evaluation function is designed for evaluating a keyword combination. This function takes into account the relationships among different keywords. Experimental results show that KCEACO can efficiently find the optimal keyword combination from a large number of candidate combinations.