Maximizing Equitable Reach and Accessibility of ETDs

William A. Ingram, Jian Wu, Edward A. Fox

TL;DR

This paper tackles the accessibility barriers of Electronic Theses and Dissertations (ETDs), which are predominantly distributed as PDFs. The authors propose using AI techniques like NLP and computer vision to convert these PDFs into machine-readable HTML, complete with semantic tags, extracted figures/tables, and generated summaries/keywords, aiming to make this scholarly output accessible to a broader audience.

What Problem Does This Paper Address?

Electronic Theses and Dissertations (ETDs) are crucial repositories of academic research, but their primary format, PDF, presents significant accessibility challenges. PDFs, while excellent for preserving visual layout, are often not machine-readable, lack semantic structure, and can be difficult to navigate for users with visual impairments, learning disabilities, or even for researchers seeking specific information within lengthy documents. This limits the equitable reach and discoverability of valuable scholarship. The paper aims to bridge this gap by making ETDs more accessible and usable for a wider range of users and systems.

Methods

The core of the proposed solution involves leveraging Artificial Intelligence (AI) to transform PDF ETDs into more accessible formats. This includes:

  • Natural Language Processing (NLP): To understand and process the textual content.
  • Computer Vision: To analyze and extract visual elements like figures and tables.
  • Text Analysis: To identify key information, generate summaries, and extract keywords.

The ultimate goal is to convert PDFs into machine-readable HTML documents that incorporate semantic tags and structure, making the content understandable to both humans and machines. This process aims to extract figures and tables and generate summaries and keywords to enhance navigation and comprehension.

Datasets

The authors mention compiling a research corpus of over 500,000 full-text ETDs and metadata from more than 40 institutional repositories across the United States. This diverse corpus is used for training models to improve their generalizability and adaptability across various writing styles, subject matters, and discourse conventions found in ETDs from different fields and disciplines.

Key Insights and Ideas

  • PDFs are a double-edged sword for scholarly communication: While preserving layout, they hinder machine readability and broad accessibility.
  • AI is key to unlocking ETD potential: NLP, computer vision, and text analysis offer powerful tools to overcome PDF limitations.
  • Beyond basic conversion: The goal isn't just to convert PDF to HTML, but to enrich it with semantic structure, extracted elements, and generated summaries for better usability.
  • Diversity in training data matters: Using a broad corpus of ETDs from various disciplines improves model robustness.
  • User-centric design is crucial: Future work needs to consider how users actually want to navigate and consume long, complex documents like ETDs.

Key Results

The paper is presented as a poster, and while it outlines a proposed approach and preliminary work, it does not present specific quantitative results from a completed system. The "results" are more in the nature of the proposed methodology and the identified need. The preliminary work involved training models for tasks such as metadata extraction, figure and table extraction, summarization, keyword generation, topic modeling, and PDF-to-XML conversion. The goal is to create a more accessible, navigable, and machine-readable DL for ETDs.

Limitations and Future Work

The authors acknowledge that creating a single, long HTML representation of an ETD might not be the optimal way for users to consume such complex documents. Future work will focus on:

  • User studies: Assessing the impact of the proposed techniques on accessibility and usability, particularly for users with visual impairments and cognitive or learning disabilities.
  • Navigation and consumption strategies: Researching the most effective ways for users to navigate and understand information within ETDs, considering their length and diverse content.
  • Hybrid approaches: Exploring combinations of structured navigation, adaptive interfaces, and summarization to help users find and understand buried content.

Summary updated

Abstract

This poster addresses accessibility issues of electronic theses and dissertations (ETDs) in digital libraries (DLs). ETDs are available primarily as PDF files, which present barriers to equitable access, especially for users with visual impairments, cognitive or learning disabilities, or for anyone needing more efficient and effective ways of finding relevant information within these long documents. We propose using AI techniques, including natural language processing (NLP), computer vision, and text analysis, to convert PDFs into machine-readable HTML documents with semantic tags and structure, extracting figures and tables, and generating summaries and keywords. Our goal is to increase the accessibility of ETDs and to make this important scholarship available to a wider audience.
BibTeX
@inproceedings{ingram2023maximizing,
  author = {William A. Ingram and Jian Wu and Edward A. Fox},
  title = {{Maximizing Equitable Reach and Accessibility of ETDs}},
  booktitle = {JCDL},
  year = {2023},
  url = {https://openreview.net/forum?id=CoJDH50Uv6},
}
Figure