Comparative Advances in Financial Sentiment Analysis:A Review of BERT,FinBert, and Large Language Models

Capturing market sentiments and supporting well-informed financial decision-making depend on the developing field of Financial Sentiment Analysis (FSA). Natural language processing (NLP) has made significant strides in comprehending and categorizing sentiment in intricate financial texts, especially with the use of Large Language Models (LLMs). The application of various LLMs, such as Bidirectional Encoder Representations form Transformers(BERT) and Financial BERT (FinBERT), as well as distilled models, such as DistilBERT and DistilRoBERTa, on a variety of financial datasets, including Financial Phrase Bank and LexisNexis news articles, was the main focus of this review article. highlighting different approaches such as model fine-tuning, zero-shot and few-shot learning, and prompt engineering. The study focuses on practical model predictions through a case study of sentimental analysis of cryptocurrencies. While FinBERT, a financial variant of BERT, exhibits high accuracy and robustness, other LLMs exhibit varying degrees of success based on the dataset and domain requirements. The analysis concentrates on the difficulties, compromises, and potential paths for improving LLMs for financial sentiment analysis.