This thesis investigates the application of deep learning architectures in natural language processing, specifically examining how transformer-based models can improve sentiment analysis accuracy in multilingual contexts. The research addresses a critical gap in cross-lingual transfer learning methodologies.