Unlocking Sentiment Analysis Potential for Low-Resource Bangla Language by Fine-Tuning LLaMA Model

Sentiment analysis, which involves the identification and assessment of the emotional state of the author, constitutes an integral component within the broader context of text analysis. This approach is crucial since it facilitates a holistic understanding of users’ emotions, perspectives, and preferences. The advent of large language models (LLMs), exemplified by LLaMA, has significantly expanded the accessibility of state-of-the-art model applications, including sentiment analysis. Nevertheless, the potential of LLMs in addressing various sentiment analysis challenges has not been extensively explored. This work explores the feasibility of leveraging LLaMA, a Large Language Model (LLM), for sentiment analysis in Bangla, a low-resource language. A dataset of 1,000 Bangla customer reviews was used to fine-tune the model, achieving an overall accuracy of 89.76%, outperforming existing models such as Bangla-BERT by 1%. The study highlights the use of parameter-efficient fine-tuning techniques (LORA and PEFT) to reduce computational overhead, making it suitable for resource-constrained environments. The findings demonstrate the potential of LLMs in advancing sentiment analysis for low-resource languages.