Search for: All records

Short answer assessment is a vital component of science education, allowing evaluation of students' complex three-dimensional understanding. Large language models (LLMs) that possess human-like ability in linguistic tasks are increasingly popular in assisting human graders to reduce their workload. However, LLMs' limitations in domain knowledge restrict their understanding in task-specific requirements and hinder their ability to achieve satisfactory performance. Retrieval-augmented generation (RAG) emerges as a promising solution by enabling LLMs to access relevant domain-specific knowledge during assessment. In this work, we propose an adaptive RAG framework for automated grading that dynamically retrieves and incorporates domain-specific knowledge based on the question and student answer context. Our approach combines semantic search and curated educational sources to retrieve valuable reference materials. Experimental results in a science education dataset demonstrate that our system achieves an improvement in grading accuracy compared to baseline LLM approaches. The findings suggest that RAG-enhanced grading systems can serve as reliable support with efficient performance gains. 

Open-text responses provide researchers and educators with rich, nuanced insights that multiple-choice questions cannot capture. When reliably assessed, such responses have the potential to enhance teaching and learning. However, scaling and consistently capturing these nuances remain significant challenges, limiting the widespread use of open-text questions in educational research and assessments. In this paper, we introduce and evaluate GradeOpt, a unified multiagent automatic short-answer grading (ASAG) framework that leverages large language models (LLMs) as graders for short-answer responses. More importantly, GradeOpt incorporates two additional LLM-based agents—the reflector and the refiner—into the multi-agent system. This enables GradeOpt to automatically optimize the original grading guidelines by performing self-reflection on its errors. To assess GradeOpt's effectiveness, we conducted experiments on two representative ASAG datasets, which include items designed to capture key aspects of teachers' pedagogical knowledge and students' learning progress. Our results demonstrate that GradeOpt consistently outperforms representative baselines in both grading accuracy and alignment with human evaluators across different knowledge domains. Finally, comprehensive ablation studies validate the contributions of GradeOpt's individual components, confirming their impact on overall performance. 

Abstract We discuss transforming STEM education using three aspects: learning progressions (LPs), constructed response performance assessments, and artificial intelligence (AI). Using LPs to inform instruction, curriculum, and assessment design helps foster students’ ability to apply content and practices to explain phenomena, which reflects deeper science understanding. To measure the progress along these LPs, performance assessments combining elements of disciplinary ideas, crosscutting concepts and practices are needed. However, these tasks are time-consuming and expensive to score and provide feedback for. Artificial intelligence (AI) allows to validate the LPs and evaluate performance assessments for many students quickly and efficiently. The evaluation provides a report describing student progress along LP and the supports needed to attain a higher LP level. We suggest using unsupervised, semi-supervised ML and generative AI (GAI) at early LP validation stages to identify relevant proficiency patterns and start building an LP. We further suggest employing supervised ML and GAI for developing targeted LP-aligned performance assessment for more accurate performance diagnosis at advanced LP validation stages. Finally, we discuss employing AI for designing automatic feedback systems for providing personalized feedback to students and helping teachers implement LP-based learning. We discuss the challenges of realizing these tasks and propose future research avenues. 

The Framework for K-12 Science Education recognizes modeling as an essential practice for building deep understanding of science. Modeling assessments should measure the ability to integrate Disciplinary Core Ideas and Crosscutting Concepts. Machine learning (ML) has been utilized to score and provide feedback on open-ended Learning Progression (LP)-aligned assessments. Analytic rubrics have been shown to be easier to evaluate the validity of ML-based scores. A possible drawback of using analytic rubrics is the potential for oversimplification of integrated ideas. We demonstrate the deconstruction of a 3D holistic rubric for modeling assessments aligned LP for Physical Science. We describe deconstructing this rubric into analytic categories for ML training and to preserve its 3D nature. 

Abstract Understanding the world around us is a growing necessity for the whole public, as citizens are required to make informed decisions in their everyday lives about complex issues. Systems thinking (ST) is a promising approach for developing solutions to various problems that society faces and has been acknowledged as a crosscutting concept that should be integrated across educational science disciplines. However, studies show that engaging students in ST is challenging, especially concerning aspects like change over time and feedback. Using computational system models and a system dynamics approach can support students in overcoming these challenges when making sense of complex phenomena. In this paper, we describe an empirical study that examines how 10th grade students engage in aspects of ST through computational system modeling as part of a Next Generation Science Standards-aligned project-based learning unit on chemical kinetics. We show students’ increased capacity to explain the underlying mechanism of the phenomenon in terms of change over time that goes beyond linear causal relationships. However, student models and their accompanying explanations were limited in scope as students did not address feedback mechanisms as part of their modeling and explanations. In addition, we describe specific challenges students encountered when evaluating and revising models. In particular, we show epistemological barriers to fruitful use of real-world data for model revision. Our findings provide insights into the opportunities of a system dynamics approach and the challenges that remain in supporting students to make sense of complex phenomena and nonlinear mechanisms. 

Involving students in scientific modeling practice is one of the most effective approaches to achieving the next generation science education learning goals. Given the complexity and multirepresentational features of scientific models, scoring student-developed models is time- and cost-intensive, remaining one of the most challenging assessment practices for science education. More importantly, teachers who rely on timely feedback to plan and adjust instruction are reluctant to use modeling tasks because they could not provide timely feedback to learners. This study utilized machine learn- ing (ML), the most advanced artificial intelligence (AI), to develop an approach to automatically score student- drawn models and their written descriptions of those models. We developed six modeling assessment tasks for middle school students that integrate disciplinary core ideas and crosscutting concepts with the modeling practice. For each task, we asked students to draw a model and write a description of that model, which gave students with diverse backgrounds an opportunity to represent their understanding in multiple ways. We then collected student responses to the six tasks and had human experts score a subset of those responses. We used the human-scored student responses to develop ML algorithmic models (AMs) and to train the computer. Validation using new data suggests that the machine-assigned scores achieved robust agreements with human consent scores. Qualitative analysis of student-drawn models further revealed five characteristics that might impact machine scoring accuracy: Alternative expression, confusing label, inconsistent size, inconsistent position, and redundant information. We argue that these five characteristics should be considered when developing machine-scorable modeling tasks.