Comparison of Ensemble Methods for Sentiment Analysis in Serbian Language

Sentiment analysis plays a significant role in natural language processing (NLP). This paper compares different ensemble methods for Sentiment analysis in Serbian language, as a low-resource language. No previous studies have experimented with ensemble methods in Serbian language and therefore this is a novel study. Dataset used in our experiment consists of movie reviews in Serbian language. Our experiment shows that ensemble of Large Language Models (LLMs) like BERT, RoBERTa, GPT-2, DistilBERT, that represent Transformers-based architecture and traditional models can boost the accuracy of the system in comparison with the base learners mentioned.