Automating Hypertext Assignment in Audit Documents: A Large Language Model-Based Approach

Auditors routinely analyze extensive datasets drawn from contracts, financial statements, and ancillary reports. This work is slow, costly, and prone to mistakes. In 1989, Laura DeYoung introduced the concept of hypertext in auditing, which involves connecting documents with electronic links similar to those found on websites. This paper builds on that idea by using Large Language Models (LLMs), which are artificial intelligence systems trained to read and interpret text, to automatically create these links. The proposed framework employs a hypergraph model, a type of network where important items, such as company names, revenues, and compliance rules, are identified and connected across documents and to external sources, while being stored in a graph database. Inspired by the eXtensible Business Reporting Language, the framework utilizes machine-readable tags to enhance consistency and validation. While initial tests indicate feasibility, future work must address challenges related to privacy, cost, and adoption to fully realize the benefits of this approach.