Topic and sentiment analysis of Slovene news media using natural language processing

Abstract

We compare topics and sentiment in Slovenian news media. We analysed the sentiment of seven media concerning specific political events or topics in 2019 and 2020. We used two phases of LDA modelling to detect a number of specific topics. For the sentiment analysis task, we fine-tuned large pretrained Slovenian masked language model, SloBERTa, and used it to classify articles in one of three classes (positive, neutral or negative). In the set of selected topics, we observed considerable differences between media in frequency and sentiment of reporting.