Automatic identification of academic articles in Japanese PDF files

As open-access becomes common, many researchers deposit their research products in a publicly accessible web (i.e. self-archiving). Although they are accessible from general search engines, massive other contents tend to hide them. The purpose of this research is to identify academic articles or quasi-articles from the entire web automatically. In this paper we conduct experiments on the performance of various classifiers and compare in terms of precision, recall, F-value. The classifiers used such attributes as terms appeared in PDF files and empirical rules. The diverse performance of each classifier discloses its characteristics.