Aligning Educational Content with Curricula Using Large Language Models

• In Event: GenAI and Adaptive Tutoring: Insights From Case Studies and Research Findings

Abstract

Educational technology systems like Khan Academy and OATutor often need to align their content with specific curricula, catering to particular classrooms or learners. These systems typically tag content with skills or standards from taxonomies to facilitate curriculum integration and adaptive learning. However, adopting new taxonomies or aligning with different curricula requires frequent content re-tagging, which can be time-consuming and resource-intensive.
To address this challenge, we propose using large language models (LLMs) to tag educational content with relevant skills from various taxonomies automatically. Our research demonstrates that fine-tuning LLMs with a small set of labeled examples significantly improves tagging accuracy compared to previous methods. Our approach is effective even for content from countries and languages underrepresented in LLM training data, countering common findings that LLMs often exhibit strong biases towards English and US domain knowledge.
We showcase the practical applications of our method by retagging and reorganizing OATutor content to align with the Common Core Standards and a Swedish calculus course syllabus. These examples demonstrate how our approach can help educational platforms efficiently tailor their systems to specific regional and curricular needs while promoting the equitable use of AI in education.

Authors

• Yerin Kwak, University of California - Berkeley

• Yunting Liu, University of California - Berkeley

• Zachary A. Pardos, University of California - Berkeley