A systematic review of sentiment analytics in banking headlines

Abstract

This systematic review investigates sentiment analysis of news headlines in the banking sector, a field susceptible to public sentiment, as demonstrated by phenomena like bank runs leading to rapid deposit withdrawals. We trace the evolution of analytic methods from traditional machine learning to advanced deep learning models, notably Bidirectional Encoder Representations from Transformer (BERT) and Generative Pre-trained Transformer (GPT). Our study highlights their applications including headline generation, sentiment measurement, fake news detection, and analysis of political bias. Despite significant advancements, we uncover research gaps, such as the ineffective use of these methodologies in banking analysis, the underuse of GPT, and a focus on performance rather than practical application. Looking ahead, we note the increasing significance of Large Language Model (LLM), the untapped potential of headline analysis in banking, and the growing interest in this area spurred by rapid technological advancements. Our findings emphasise the pivotal role of sentiment analysis in deciphering market trends and improving decision making in finance, underscoring its strategic importance in the banking industry.