Exploration on Application of Large Language Models in Unstructured Audit Materials

Abstract

Research on the application of large language models (LLMs) and natural language processing (NLP) in unstructured audit data. Traditionally, auditors have used only the quantitative financial data, and unstructured data such as emails, contracts and meeting minutes have been difficult to process on a large scale, thus increasing audit risk. Newer LLMs can be used to combine text and numbers in order to strengthen the risk assessment and financial statement analysis work. Review of the theory and practice of AI application in unstructured audit evidence. Research has been conducted on how to use large language models (LLMs) for automated compliance checks, sentiment analysis and anomaly detection of narrative context combined with quantitative disclosures. In addition, there are technical and governance problems with the above models, such as algorithmic bias, data privacy, and the requirement for continuous monitoring. Although LLMs have enhanced the extraction and analysis of unstructured data, a strong governance system and an immutable audit trail are needed for their application to ensure professional scepticism and regulatory compliance in the new era of digital audits, as mentioned in this paper.