The future of document indexing: GPT and Donut revolutionize table of content processing

TL;DR

The paper tackles the challenge of automating the extraction of structured information from lengthy construction specification documents by focusing on the table of contents (ToC) as an indexing entry point. It combines OCR-free Visual Document Understanding with the Donut model and OpenAI GPT-3.5 Turbo prompts to automatically identify ToC pages and convert their content into a structured JSON representing headings and subheadings. The proposed end-to-end pipeline includes page classification, ToC extraction, JSON structuring, API access, and a visualization dashboard, achieving an overall accuracy of 82.2% on 20 test documents with individual component accuracies in the 79–90% range; Donut and GPT-3.5 Turbo demonstrated 85% and 89% effectiveness in structuring ToCs, respectively. This approach demonstrates the potential to dramatically improve document indexing efficiency across industries, reducing manual data extraction and enabling scalable navigation of complex specifications.

Abstract

Industrial projects rely heavily on lengthy, complex specification documents, making tedious manual extraction of structured information a major bottleneck. This paper introduces an innovative approach to automate this process, leveraging the capabilities of two cutting-edge AI models: Donut, a model that extracts information directly from scanned documents without OCR, and OpenAI GPT-3.5 Turbo, a robust large language model. The proposed methodology is initiated by acquiring the table of contents (ToCs) from construction specification documents and subsequently structuring the ToCs text into JSON data. Remarkable accuracy is achieved, with Donut reaching 85% and GPT-3.5 Turbo reaching 89% in effectively organizing the ToCs. This landmark achievement represents a significant leap forward in document indexing, demonstrating the immense potential of AI to automate information extraction tasks across diverse document types, boosting efficiency and liberating critical resources in various industries.

Figures (7)

• Figure 1: For this type of document, the first heading (h) is referred to as the heading, while the second heading (sh) is referred to as the subheading. The heading number (hn) comes before the heading title (ht). The same is true for the subheading number (shn) and title (sht).
• Figure 2: For this type of document, the division is referred to as the heading(h), while the sections are referred to as the subheading (sh). The heading number (hn) comes before the division title (ht). The same is true for the section number (she) and title (sht)
• Figure 3: The proposed approach for retrieving and structuring table of contents of a PDF document.
• Figure 4: A prompt to extract ToC text from the raw text of a PDF file.
• Figure 5: A prompt to extract key information based on a given example by formatting the output according to the provided schema.