Document classification to functional styles (Domains of use): Lithuanian case

Abstract

We report an experiment on classification of Lithuanian texts according to their domain (area of use), i.e. functional style. Functional style is a variety of standard language that is defined by domain, contents, functions, stylistic devices and linguistic means. In this paper we discuss an experiment on document classification into 3 functional styles of Lithuanian language – administrative, publicist and scientific. We compare results of 5 algorithms: Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), k-Nearest Neighbors (k-NN), Support Vector Machine (SVM) with kernel function and Naïve Bayes. We also used 8 quantitative linguistic indicators as discriminating features. For administrative style SVM was the most effective (96.5 % of texts classified correctly), for publicist style – LDA (98.9 % of texts classified correctly) and for scientific style – QDA (93.1 % of texts classified correctly). We achieved the best F-score with SVM (94.7 % – for administrative style, 98.9 – for publicist style and 85.9 – for scientific style)