Research on Financial Statement Checking Relationship Recognition System Based on Large Language Models

With the driving power of the internet wave, the auditing industry faces unprecedented challenges and opportunities. Traditional audit methods increasingly reveal weaknesses at processing vast data, which requires the application of new technologies to maximize the efficiency and accuracy of audit. As an epoch-making innovation of natural language processing, Large Language Models (LLMs) demonstrate unequalled performance at parsing texts, semantic detection, and text generation, opening up approaches to forward-looking intelligent reform of auditing. This paper researches the innovation of LLMs to be utilized to intelligent Checking relationship detection of financial statements, to enhance the efficiency and accuracy of auditing.Based on the combination of the knowledge base of audit with the technologies of Retrieval-Augmented Generation (RAG), we introduce a multi-agent system to intelligent Checking relationship detection of financial statements. LLMs can automatically identify and certify Checking relationships between financial statements, dramatically improving audit efficiency and quality. Environmental context and data support are provided through the audit knowledge base, and the RAG technologies enhance the power of analysis. This paper demonstrates, through experiments, that these technologies can support intelligent Checking relationship detection, ushering auditing into an intelligent audit era.