Sentiment Analysis for Low-Resource Languages Using a Hybrid XLM-Roberta and Bi-LSTM Model

In this study, we tackle the challenges of sentiment analysis in low-resource languages by analyzing social media datasets with natural language processing (NLP). Previous research has highlighted several limitations, including difficulties in managing informal language, limited use of emoticons, and accuracy limits in traditional models such as CNN and LSTM. Furthermore, the absence of fine-tuning for large-language models has hindered performance. Issues such as semantic complexity, translation errors, and generally poor outcomes in low-resource languages have also persisted. To address these challenges, we introduce a hybrid LLM model that combines the strengths of XLM-R and Bi-LSTM, which has been proven to deliver more accurate results compared to traditional approaches. By fine-tuning these large language models for better adaptation to specific contexts and streamlining the complexity of our methods, particularly in low-resource settings. Our findings classify sentiments into distinct categories such as “Not offensive,” “Offensive_Untargeted,” and various forms of targeted insults. Ultimately, this study not only contributes to the design of effective sentiment analysis frameworks using hybrid LLM models but also discusses their implications for business and management. Improved sentiment analysis can provide valuable insights that inform decision-making and enhance customer engagement strategies, making it a vital tool in today's data-driven world. We significantly improve multilingual performance with an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">${6 2 \%}$</tex>.