Using One-Class SVM for Scientific Documents Classification Case study: Iranian Environmental Thesis

Rabiei, Mohammad; MahdiHosseini-Motlagh, Seyyed-Mahdi; Minaei Bidgoli, Behrouz

doi:10.35050/JIPM010.2019.036

Using One-Class SVM for Scientific Documents Classification Case study: Iranian Environmental Thesis

Authors

Mohammad Rabiei

Seyyed-Mahdi MahdiHosseini-Motlagh

Behrouz Minaei Bidgoli

10.35050/JIPM010.2019.036

Abstract

The classification of research studies is important in order to identify and analyze the research supply and demand in various fields of science. In particular, the classification of environmental research is essential because of its importance in Iran and its interdisciplinary nature. This research proposes One-Class Classification (OCC) method to classify the research studies in this domain using Support Vector Machine (SVM) and consequently evaluates important parameters affecting the quality of this classification. The results show that the use of descriptive metadata has better performance than the content metadata in order to make a core data set to learn the model. Moreover, the use of the polynomial kernel and the binary weighing of words in the features vector matrix leads to better results than other states. In this paper a new weighing method has been proposed which is superior to the other methods especially in precision criterion. We call this weighing method as NG-TF, which can be used in term-document matrix to determine the indicator terms of scientific domains.

Keywords

environment

One-Class Classification

Support Vector Machine (SVM)

text mining

NG-TF Weighting