Conference Paper

Exploring NLP-based Methods for Generating Engineering Ethics Assessment Qualitative Codebooks

A. Hingle, A. Katz, A. Johri

IEEE Frontiers in Education Conference (FIE)2023

Contribution Summary

The paper offers an early comparison of assisted and more automated NLP workflows for qualitative codebook generation in engineering ethics assessment.

Draft enrichment generated from extracted publication text; pending human review.

Plain-Language Summary

This paper compares two NLP-supported approaches for generating qualitative codebooks from engineering ethics reflection data. It uses student responses from a technology ethics course to examine how human-NLP collaboration and more automated NLP workflows produce similar and different themes.

Research Question

How effective are NLP-supported methods for generating a qualitative codebook for student reflections on technology ethics?

Methods

  • Analyzed open-ended student responses collected across six iterations of a semester-long technology ethics course.
  • Compared a Human-NLP workflow, where researchers revised model-generated themes, with an Auto-NLP workflow using iterative embedding, clustering, and summarization.
  • Used Python, sentence-transformer embeddings, agglomerative clustering, and an LLM summarization process to generate candidate themes.

Key Findings

  • The Human-NLP method produced eight final themes grouped into three overarching themes.
  • The Auto-NLP method produced twelve final themes grouped into four overarching themes, with substantial overlap across methods.
  • Both approaches highlighted students' learning about ethics, technology-society connections, and organizational responsibility, while also showing risks around context loss and non-determinism.

Implications

NLP can help researchers make sense of large text corpora, but generated codebooks need researcher review before use.

Method choices such as clustering algorithms, model settings, and human review steps materially affect qualitative outputs.

Engineering ethics assessment can benefit from scalable qualitative workflows that preserve attention to course context and student meaning.

Research Artifacts

protocolHuman-NLP codebook generation workflowA human-in-the-loop workflow for generating, reviewing, and revising qualitative themes.
protocolAuto-NLP codebook generation workflowAn iterative embedding, clustering, and summarization workflow for automated candidate codebook generation.

Abstract

Publication on Exploring NLP-based Methods for Generating Engineering Ethics Assessment Qualitative Codebooks

Related Projects