Analyzing the characteristics of academic paper categories by using an index of representativeness

Analyzing the characteristics of academic paper categories by using an index of representativeness ⋆ Takafumi Suzuki a, Kiyoko Uchiyama b, Ryota Tomisaka c, and Akiko Aizawa b aFaculity of Sociology, Toyo University, 5-28-20, Hakusan, Bunkyo-ku, Tokyo 112-8606, Japan takafumi s@toyo.jp bDigital Content and Media Sciences Research Division, National Institute of Informatics 2-1-2, Hitotsubashi, Chiyoda-ku, Tokyo 101-8430, Japan {kiyoko, aizawa}@nii.ac.jp cGraduate School of Information Science and Technology, The University of Tokyo 7-3-1, Hongo, Bunkyo-ku, Tokyo 112-0023, Japan ryota.tomisaka@gmail.com Abstract. This study proposes an index of representativeness for analyzing the characteris- tics of academic paper categories. Many textual indices have been proposed in the field of computational stylistics, but all of the previous indices are limited in that they (a) focus only on the styles of the texts; (b) return an absolute value for every text, and (c) are based on the number of tokens. In this study, we propose an index of representativeness that does not have the weaknesses of the previous indices. Our index is based on the h-index that was originally proposed in the field of scientometrics. We redefine it here for textual data. We show the effectiveness of our index for analyzing the characteristics that differ between four genres and three subfields in Japanese academic papers. Keywords: academic papers, h-index, lexical indicators 1 Introduction Many textual indices have been proposed in the field of computational stylistics. The number of tokens, i.e., the frequency of words in a document, is the simplest. Some indices use information related to words that appear only one or two times in the text, while others take the entire frequency spectrum into account (Kageura, 2000; Tweedie and Baayen, 1998). Textual indices have been studied for various applications such as authorship analysis, genre detection, and chronological text mining (Argamon et al., 2007; Koppel et al., 2009; Stamatatos, 2009), but they have three flaws in common: (a) they shed light only on the styles of text; (b) they return an absolute value, meaning that they do not consider the relative values for different keyword sets or the relative position of a document in the document set; and (c) they are based on the number of tokens. Our problem is the use of textual indices to analyze the characteristics of the categories of academic papers. Since a wide variety of texts are available, these types of the ‘characterization’ problems, as well as the ‘classification’ problems, are currently increasing in importance. To deal with the characterization problems, we aim at deriving a meaningful level of knowledge ⋆ This study was supported by Grant-in-Aid for Scientific Research 21800087 for Young Scientists (Start-up) and 23700288 for Young Scientists (B), from the Ministry of Education, Culture, Sports, Sciences and Technology, Japan, and Grant for joint research from National Institute of Informatics. This study used the text data provided by the Information Processing Society of Japan (IPSJ). Earlier version of this study was presented at 16th Annual Meeting of the Association for Natural Language Processing (NLP2010). We would like to thank all …

Content truncated to 500 words