Abstract:

Abstract The paper presents the compilation of the Trendi corpus, the first monitor corpus of Slovene. The current version (Trendi 2023-02) contains texts published between January 2019 and October 2023, with a total of over 700 million tokens (more than 586 million words). The purpose of the corpus is to provide linguists and non-linguists with data on current language use and to enable the monitoring of new words as well as the increase and decline in the use of existing words. In the paper, we present the contents of the corpus and the methods and criteria used in its compilation. The second part of the paper is focused on the development of a tool for categorizing text topics in news articles. The tool was developed specifically for the Trendi corpus but can be used for other corpora containing similar texts. A set of 13 thematic categories was developed for the tool. The set generally follows international standards and categories used in comparable corpora for other languages. Using texts annotated with these categories, we trained multiple language models and achieved a high classification accuracy when categorizing text topics.

Keywords: