A comparative study of cross-lingual sentiment analysis

Open Webpage

Pavel Pribán, Jakub Smíd, Josef Steinberger, Adam Mistera

Published: 2024, Last Modified: 17 Jul 2025Expert Syst. Appl. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•XLM-R model outperforms other approaches in cross-lingual sentiment analysis.•Linear transformation achieves similar results to smaller Transformer models.•Linear transformation can be used as a faster alternative.•We provide a set of recommendations for linear transformation hyper-parameters.•ChatGPT achieves similar or better results than smaller models in zero-shot setting.