2026 ASEE Annual Conference & Exposition

Examining Accuracy in AI-Powered Metadata Extraction for Engineering Education Research

Presented at Engineering Libraries Division (ELD) Poster Session

Examining Accuracy in AI-Powered Metadata Extraction for Engineering Education Research

Abstract

Background: Engineering libraries are drowning in paperwork. Every time a researcher publishes a paper, someone has to record key details about it: who wrote it, where they work, and what the paper is about. As the volume of engineering education research grows, doing this by hand is no longer realistic. Many libraries are turning to automated tools to do this job, but no one has tested how well these tools actually work or where they fail.
Purpose: This study asks a simple but urgent question: which automated tool does the best job of pulling key information from research papers, and can a smarter combination of tools do better than any single one alone?
Method: We tested four different automated approaches on 50 peer-reviewed engineering education papers, comparing each tool's output against information that was verified by humans. The four tools ranged from a free, widely used academic document reader, to a rule-based system, to a modern AI assistant, to a custom hybrid approach that assigns each task to whichever tool handles it best.
Results: The hybrid approach, AlphaMind-Extrator, outperformed all others. More importantly, the study uncovered a critical blind spot: one of the most popular free tools used in libraries today gets author affiliations, which university or department an author belongs to, wrong more than 96% of the time, while appearing reliable on everything else. The hybrid approach fixed this failure, achieving 86% accuracy on affiliations at a cost of less than two cents for 50 papers.
Conclusions: No single automated tool does everything well. The safest and most cost-effective approach is to match each task to the tool best suited for it, and to test tools field by field before trusting them with your data.

Authors
  1. Mr. Joshua Owusu Ansah Arizona State University [biography]
  2. Dr. Brooke Charae Coley Arizona State University, Polytechnic Campus [biography]
  3. Monica Lynn Miles Orcid 16x16http://orcid.org/0000-0003-0006-1842 University at Buffalo, The State University of New York [biography]
Note

The full paper will be available to logged in and registered conference attendees once the conference starts on June 21, 2026, and to all visitors after the conference ends on June 24, 2026