Leveraging Large Language Models for Economic and Management Research: BERT-Based Sentiment Analysis

Abstract

Large language models (LLMs) are fundamentally reshaping the landscape of unstructured text data analysis. However, their application in the field of economic and management research in China remains constrained. Given the challenges encountered in sentiment analysis of financial text, this study utilizes the BERT model as an example to elucidate the potential applications and methods of LLMs within economic and management domains. Specifically, this paper employs BERT for sentiment classification, introduces four novel financial sentiment measures, and comprehensively assesses their effectiveness in terms of classification performance, explanatory power, and real economic value. The results demonstrate that the FinBERT model excels in measuring financial text sentiment, exhibiting significant advantages compared to traditional dictionary methods. This paper provides a potent tool for enhancing financial text sentiment analysis, with substantial practical implications for advancing capital market efficiency and strengthening non-financial information regulation. It also introduces LLMs to the economics and management field, offering valuable insights for prospective research endeavors involving such models.

Keywords: financial text; sentiment analysis; large language model; deep learning

JEL Classification: M4,G3