Automating Chapter-Level Classification for Electronic Theses and Dissertations
Project Overview
The document explores the application of generative AI, specifically machine learning and language models, in the education sector by focusing on the automation of chapter-level classification for electronic theses and dissertations (ETDs). It highlights the inadequacy of traditional high-level metadata in accurately representing the complexity of ETDs, which hampers their discoverability. By leveraging an AI-driven approach to generate detailed chapter-level metadata, the research aims to enhance accessibility, improve information retrieval, and facilitate interdisciplinary research. The paper evaluates different classification strategies to identify the most effective methods for categorizing ETD chapters, ultimately demonstrating how generative AI can significantly improve the organization and accessibility of academic resources, thereby supporting better educational outcomes and research collaboration.
Key Applications
Automated chapter-level classification for ETDs
Context: Used in academic libraries for improving discoverability of electronic theses and dissertations
Implementation: Utilized machine learning and AI to segment ETDs into chapters and classify them with detailed metadata
Outcomes: Improved access to specific sections of ETDs, enhanced discoverability for interdisciplinary research, and better support for academic inquiries.
Challenges: Complexity of accurately segmenting chapters, variations in formatting across disciplines, and the need for subject matter expertise to validate AI-generated classifications.
Implementation Barriers
Technical Barrier
Challenges associated with segmenting ETDs into individual chapters due to varying formats and lack of support in PDFs.
Proposed Solutions: Combining AWS Textract with object detection techniques to improve text extraction and chapter segmentation.
Resource Barrier
The need for subject matter expertise to evaluate the accuracy of AI-generated classifications, which can be costly and time-consuming. This includes implementing robust prompting strategies and continuous feedback mechanisms to refine AI outputs.
Proposed Solutions: Developing strategies to lower the costs and time associated with obtaining subject matter expertise.
Project Team
Bipasha Banerjee
Researcher
William A. Ingram
Researcher
Edward A. Fox
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Bipasha Banerjee, William A. Ingram, Edward A. Fox
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai