A Meta-Learning Augmented Bidirectional Transformer Model for Automatic Short Answer Grading
Conference Paper
Wang, Z., Lan, A. S., Waters, A. E., Grimaldi, P., & Baraniuk, R. G. (2019, July). In EDM.
We introduce ml-BERT, an effective machine learning method for automatic short answer grading when training data, i.e., graded answers, is limited. Our method combines BERT (Bidirectional Representation of the Transformer), the state-of-the-art model for learning textual data representations, with meta-learning, a training framework that leverages additional data and learning tasks to improve model performance when labeled data is limited. Our intuition is to use meta-learning to help us learn an initialization of the BERT parameters in a specific target subject domain using unlabeled data, thus fully leveraging the limited labeled training data for the grading task.
Towards Blooms Taxonomy Classification Without Labels
Conference Paper
Wang, Z., Manning, K., Mallick, D. B., & Baraniuk, R. G. (2021, June). In International Conference on Artificial Intelligence in Education (pp. 433-445). Springer, Cham. In this work, we explore weakly supervised machine learning for classifying questions into distinct Bloom’s Taxonomy levels. Bloom’s levels provide important information that guides teachers and adaptive learning algorithms in selecting appropriate questions for their students. However, manually providing Bloom labels is expensive and labor-intensive, which motivates a machine learning approach. Current automated Bloom’s level classification methods employ supervised learning that relies on large, labeled datasets that are difficult and costly to construct. In this paper, we propose a weakly supervised learning method that performs binary Bloom’s level labeling without any a priori known Bloom’s taxonomy labels.
Open-Ended Knowledge Tracing
Pre-print
Liu, N., Wang, Z., Baraniuk, R. G., & Lan, A. (2022). In this paper, we conduct the first exploration into open-ended knowledge tracing (OKT) by studying the new task of predicting students’ exact open-ended responses to questions. Our work is grounded in the domain of computer science education with programming questions. We develop an initial solution to the OKT problem, a student knowledge-guided code generation approach, that combines program synthesis methods using language models with student knowledge tracing methods. We also conduct a series of quantitative and qualitative experiments on a real-world student code dataset to validate OKT and demonstrate its promise in educational applications.
Towards Human-like Educational Question Generation with Large Language Models
Conference Paper
Wang, Z., Valdez, J., Basu Mallick, D., Baraniuk, R. (2022). At the Artificial Intelligence in Education (AIED) conference. Durham, UK. We investigate the utility of large pretrained language models (PLMs) for automatic educational assessment question generation. In this paper, we investigate the impact of various PLM prompting strategies on the quality of generated questions. We design a series of generation scenarios to evaluate various generation strategies and evaluate generated questions via automatic metrics and manual examination. With empirical evaluation, we identify the prompting strategy that is most likely to lead to high-quality generated questions. Finally, we demonstrate the promising educational utility of generated questions using our concluded best generation strategy by presenting generated questions together with human-authored questions to a subject matter expert, who despite their expertise, could not effectively distinguish between generated and human-authored questions.
Mathematics Teachers’ Use of Generative AI to Create Active Learning Experiences
Journal
Walkington, C., & Bainbridge, K. (2025). Social Innovations Journal, 30(2).
Generative AI tools have arisen as important supports to assist teachers with planning instruction. One important way that GenAI can assist teachers in lesson planning is by allowing them to incorporate more “active learning” into their instruction, which we refer to here as “generative learning.” In the present chapter, we explore mathematics teachers (n=16) using GenAI to adapt a task they are using in their classroom to incorporate generative learning and also explore GenAI as support as teachers themselves engage in a novel generative learning task related to lesson design. We find that teachers appreciate the creative affordances of generative AI to support generative learning activities but worry about time efficiency, usability, and accuracy, as well as a variety of important big-picture issues.
