大语言模型与经济金融文本分析:基本原理、应用场景与研究展望  

Text Analysis in Economics and Finance with Large Language Models:Fundamentals,Applications,and Future Prospects

作　　者：林建浩[1] 孙乐轩 LIN Jianhao;SUN Lexuan(Lingnan College,Sun Yat-sen University,Guangzhou 510275,China)

机构地区：[1]中山大学岭南学院,广州510275

出　　处：《计量经济学报》2025年第1期1-34,共34页China Journal of Econometrics

基　　金：国家社会科学基金(22AZD121);国家自然科学基金(71991474,72073148,72273156,72303258)。

摘　　要：大语言模型(LLMs)拥有强大的自然语言处理能力,本文系统梳理了相关领域的前沿文献,并详细论述其为经济金融领域的文本分析带来的新研究机会。首先,本文介绍了GPT和BERT这两类最具代表性的LLMs及一系列经济金融领域专用的LLMs,并阐述应用LLMs分析文本数据的基本原理.其次,本文从两个方面归纳了LLMs在经济金融文本分析研究中的应用场景:一是LLMs在计算文本相似度、提取文本向量进行预测、文本数据识别与分类、构建特定领域词典、主题建模与分析、文本情绪分析等传统文本分析场景中所呈现的显著优势;二是LLMs拥有强大的人类对齐能力,由此开创了全新的应用场景,即作为经济代理人,模拟人类生成对于文本的信念或预期并进行经济决策.最后,本文总结了LLMs在开拓经济金融文本分析研究的新范式过程中面临的局限性以及现有研究不足,并针对这些问题可能催生的新研究议题进行展望。Large language models(LLMs)haye powerful natural language processing capabilities.In this paper,we systematically review the recent literature in this field and highlight the new research opportunities that LLMs bring to text analysis in economics and finance.First,we introduce GPT and BERT,the two most representative LLMs,as well as a number of LLMs developed specifically for economic and financial applications.Additionally,we also elaborate on the fundamental principles behind applying LLMs for text data analysis.Second,we summarize the applications of LLMs in economic and financial text analysis from two perspectives.On the one hand,we highlight the significant advantages of LLMs in traditional text analysis scenarios,such as calculating text similarity,extracting text vectors for prediction,text data identification and classification,building domain-specific dictionaries,topic modeling and analysis,and text sentiment analysis.On the other hand,LLMs have strong human alignment capabilities,thus opening up entirely new application scenarios,i.e.,acting as economic agents that simulate humans in generating beliefs or expectations about texts and making economic decisions.Finally,we summarize the limitations and existing research gaps that LLMs face in pioneering new paradigms of economic and financial text analysis research,and discuss potential new research topics that may arise from these issues.

关 键 词：大语言模型 自然语言处理 文本分析 经济与金融 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.1[自动化与计算机技术—控制科学与工程] F830[经济管理—金融学]