Conference Session

ERM Technical Session: Improving Assessment in Engineering Education

Collection

2025 ASEE Annual Conference & Exposition

Authors

David Coulter Jangraw, University of Vermont; Anneliese Marie Shoudt; Courtney D Giles, University of Vermont

Tagged Topics

Diversity

Tagged Divisions

Educational Research and Methods Division (ERM)

hypotheses for future research.A preliminary qualitative analysis of Ticket Home student exit surveys and TA Ticket Homesummaries used ChatGPT to identify common themes. The surveys were first manuallyanonymized by removing names, then entered into ChatGPT with the prompt “Please summarizethese responses to the question: . List the most common themes and how manyresponses mentioned each of them.” While a similar approach has been used successfully forthematic analysis before [20], our approach involves different data and prompts; it shouldtherefore be considered preliminary and subject to a more extensive validation. To assess theapproximate accuracy of this approach, a human rater manually identified the four most commoncodes identified by ChatGPT

Conference Session

Educational Research and Methods Division (ERM) Technical Session 7

Collection

2024 ASEE Annual Conference & Exposition

Authors

Xiuhao Ding, University of Illinois at Urbana - Champaign; Meghana Gopannagari, University of Illinois at Urbana - Champaign; Kang Sun, University of Illinois at Urbana - Champaign; Alan Tao, University of Illinois at Urbana - Champaign; Delu Louis Zhao; Sujit Varadhan, University of Illinois at Urbana - Champaign; Bobbi Lee Battleson Hardy, University of Illinois at Urbana - Champaign; David Dalpiaz, University of Illinois at Urbana - Champaign; Chrysafis Vogiatzis, University of Illinois at Urbana - Champaign; Lawrence Angrave, University of Illinois at Urbana - Champaign; Hongye Liu, University of Illinois at Urbana - Champaign

Tagged Topics

Diversity

Tagged Divisions

Educational Research and Methods Division (ERM)

gold standard to evaluateautomated text analytic approaches. Raw text from open-ended questions was converted intonumerical vectors using text vectorization and word embeddings and an unsupervised analysisusing document clustering and topic modeling was performed using LDA and BERT methods. Inaddition to conventional machine learning models, multiple pre-trained open-sourced local LLMswere evaluated (BART and LLaMA) for summarization. The remote online ChatGPTclosed-model services by OpenAI (ChatGPT-3.5 and ChatGPT-4) were excluded due to subjectdata privacy concerns. By comparing the accuracy, recall, and depth of thematic insights derived,we evaluated how effectively the method based on each model categorized and summarizedstudents

Conference Session

Educational Research and Methods Division (ERM) Technical Session 18

Collection

2024 ASEE Annual Conference & Exposition

Authors

Navarun Gupta, University of Bridgeport; Junling Hu, University of Bridgeport; Ioana A. Badara, Post University; Buket D. Barkana, The University of Akron; Deana A. DiLuggo, University of Bridgeport

Tagged Topics

Diversity

Tagged Divisions

Educational Research and Methods Division (ERM)

they can right away see being applied through concepts ofsimple Calculus and Python programming.Deep Convolution-based networks with the Triplet loss were quite successful (e.g., FaceNet) inface recognition, resulting in greater than 99% accuracy on benchmarks such as LFW. With therecent success of transformer-based Natural Language Processing architectures (e.g., ChatGPT),transformers have been attempted in Computer Vision applications. They have shown considerablesuccess with better computational efficiency than CNN-based architectures. In this project, wecompared the FaceNet and transformer-based architecture for face recognition. We also providedan insightful understanding of the face recognition process, its limitations, and future