Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enriching Historical Records: An OCR and AI-Driven Approach for Database Integration

Zahra Abedi, Richard M. K. van Dijk, Gijs Wijnholds, Tessa Verhoef

TL;DR

The paper describes a three-phase pipeline to enrich historical records from the Leidse hoogleraren en lectoren 1575–1815 by integrating OCR, AI-driven information extraction, and record linkage into a central LUCD database. It combines Tesseract OCR with Dutch customization and image preprocessing, a Pydantic-based JSON schema, and GPT-3.5 Turbo function calling to produce structured JSON that is then linked to existing records. Quantitative evaluations show OCR CER of 1.08% and WER of 5.06%, JSON extraction accuracy around 65% with correct text and 63% with OCR-derived text, and linking accuracy of about 94% with correct JSON and 81% with OCR-derived JSON, highlighting OCR quality as a key factor for downstream performance. The study discusses challenges from layout variability and terminological differences, and outlines future directions including multimodal AI models and volume-specific schemas to further improve accuracy and scalability in digital humanities data integration.

Abstract

This research digitizes and analyzes the Leidse hoogleraren en lectoren 1575-1815 books written between 1983 and 1985, which contain biographic data about professors and curators of Leiden University. It addresses the central question: how can we design an automated pipeline that integrates OCR, LLM-based interpretation, and database linking to harmonize data from historical document images with existing high-quality database records? We applied OCR techniques, generative AI decoding constraints that structure data extraction, and database linkage methods to process typewritten historical records into a digital format. OCR achieved a Character Error Rate (CER) of 1.08 percent and a Word Error Rate (WER) of 5.06 percent, while JSON extraction from OCR text achieved an average accuracy of 63 percent and, based on annotated OCR, 65 percent. This indicates that generative AI somewhat corrects low OCR performance. Our record linkage algorithm linked annotated JSON files with 94% accuracy and OCR-derived JSON files with 81%. This study contributes to digital humanities research by offering an automated pipeline for interpreting digitized historical documents, addressing challenges like layout variability and terminology differences, and exploring the applicability and strength of an advanced generative AI model.

Enriching Historical Records: An OCR and AI-Driven Approach for Database Integration

TL;DR

The paper describes a three-phase pipeline to enrich historical records from the Leidse hoogleraren en lectoren 1575–1815 by integrating OCR, AI-driven information extraction, and record linkage into a central LUCD database. It combines Tesseract OCR with Dutch customization and image preprocessing, a Pydantic-based JSON schema, and GPT-3.5 Turbo function calling to produce structured JSON that is then linked to existing records. Quantitative evaluations show OCR CER of 1.08% and WER of 5.06%, JSON extraction accuracy around 65% with correct text and 63% with OCR-derived text, and linking accuracy of about 94% with correct JSON and 81% with OCR-derived JSON, highlighting OCR quality as a key factor for downstream performance. The study discusses challenges from layout variability and terminological differences, and outlines future directions including multimodal AI models and volume-specific schemas to further improve accuracy and scalability in digital humanities data integration.

Abstract

This research digitizes and analyzes the Leidse hoogleraren en lectoren 1575-1815 books written between 1983 and 1985, which contain biographic data about professors and curators of Leiden University. It addresses the central question: how can we design an automated pipeline that integrates OCR, LLM-based interpretation, and database linking to harmonize data from historical document images with existing high-quality database records? We applied OCR techniques, generative AI decoding constraints that structure data extraction, and database linkage methods to process typewritten historical records into a digital format. OCR achieved a Character Error Rate (CER) of 1.08 percent and a Word Error Rate (WER) of 5.06 percent, while JSON extraction from OCR text achieved an average accuracy of 63 percent and, based on annotated OCR, 65 percent. This indicates that generative AI somewhat corrects low OCR performance. Our record linkage algorithm linked annotated JSON files with 94% accuracy and OCR-derived JSON files with 81%. This study contributes to digital humanities research by offering an automated pipeline for interpreting digitized historical documents, addressing challenges like layout variability and terminology differences, and exploring the applicability and strength of an advanced generative AI model.
Paper Structure (25 sections, 2 figures, 5 tables)

This paper contains 25 sections, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Research Questions
  4. Related Work
  5. Optical Character Recognition
  6. Information Extraction
  7. Record Linkage
  8. Data
  9. Methods
  10. Phase 1: PDF to PNG Conversion, Image Preprocessing, and OCR & Text Segmentation Per Person
  11. Phase 2: JSON Extraction from Text with AI
  12. Schema Definition
  13. Extraction Techniques
  14. Phase 3: Record Linkage & Database Enrichment
  15. Linking Algorithm
  16. ...and 10 more sections

Figures (2)

  • Figure 1: Comparison of Average Word Error Rate (WER) and Character Error Rate (CER) Across Different Volumes
  • Figure 2: Correct interpretations per category using ChatGPT-3.5 in five independent trails. The chart displays the accuracy min-max ranges and means for JSON files generated from both correct text files and OCR-generated text files across various categories.