Abstract: This study surveys recent advancements in sentiment analysis of financial review texts, both domestically and internationally, to delineate the field’s developmental trajectory. Adopting dual perspectives of technology-driven and content-driven approaches, it scrutinizes prevailing research trends. Technologically, the evolution from lexicon-based methods, through traditional machine learning, to deep learning paradigms is summarized. Content-wise, BERTopic and LLaMA3 are employed for document clustering based on scholarly viewpoints, with dynamic topic modeling elucidating domain progress. Findings indicate a domestic transition from sentiment analysis methods to investigations of emotional impacts on financial market prediction. Meanwhile, international research continues progressing deep learning applications while revealing emerging interests in financial sentiment modeling. By integrating these observations, the paper proposes future directions including: (1)constructing high-quality datasets, (2)conducting granular sentiment analysis of financial discourse, and (3)improving the interpretability of analytical outcomes. These recommendations aim to establish methodological foundations for subsequent studies in this field.

Key words: Financial reviews, Sentiment analysis, Research trends, BERTopic, LLaMA3