My name is Neha and I am a Software Development Engineer at Sonos, Inc. and a Ph.D. Candidate in the ECE department at the University of Washington. I am an interdisciplinary researcher with expertise in Natural Language Processing, Machine Learning, and Engineering Education Research, with a passion for applying data-driven approaches to solve complex problems.
Learn about what I doTextual survey data is processed manually, and researchers must invest significant time and effort in evaluating such text survey collections. While in-depth manual research of text survey data yields significant outcomes, some instructors or administrators may benefit from an automated system that can uncover trends and offer broader overviews for selected datasets in less time and resource-efficient ways. This study investigates using Natural Language Processing (Topic modeling: Latent Dirichlet Allocation and Non-negative Matrix Factorization) techniques to extract labels from student short answer survey responses in the context of Engineering. I also conducted extensive research on ethical considerations associated with the use of Natural Language Processing in Engineering Education research.
As described by Burdick et al. [2021], in this project we analyzed the stability of word embeddings across various languages. Here, stability is defined as the degree to which two embedding spaces are in perfect agreement. A stability of 100 percent shows complete agreement between the two embedding spaces, whereas a stability of zero percent suggests complete disagreement.
A single dense output layer and a multi-layer dense output layer model is used to predict peer support labels based on a cooperative learning framework. Sentiment analysis is also conducted to determine students' feelings toward faculty support using NLPTK, Textblob, and Flair.
Machine learning methods are used to forecast students' academic support and course level engagement in both traditional and remote learning environments. The Ordinary Least Squares method of linear regression is used as a baseline for comparison with other models (Ridge, Lasso, Elastic, GLR Poisson’s, GLR Gamma, Epsilon SVM, Nu SVM, Stochastic Gradient Descent). The mean square error for the test data serves as the evaluation criterion.
This research collected over 1,000 surveys and used a convergent parallel mixed-methods approach to investigate peer support in both remote and in-person settings. An independent sample t-test and Mann Whitney test conducted between all courses showed that there was no significant difference in perceived peer support between remote and in-person learning environments. Analyses of existing peer support were supplemented by qualitative analysis of short answer questions regarding student expectations for peer support. Short answer questions were deductively coded according to a cooperative learning framework.