Classification of physics problems as a basis for the development of educational AI models

Abstract

The integration of artificial intelligence (AI) into education has opened new possibilities for personalized learning, adaptive assessment, and intelligent content delivery (Tapalova & Zhiyenbayeva, 2022). However, the effectiveness of such systems largely depends on the availability of high-quality, well-structured datasets that reflect the complexity of real educational content. In the domain of science education, and particularly physics, problem-solving tasks represent a core component of learning and assessment (Küchemann et al., 2024). Despite their importance, these tasks are rarely prepared in formats suitable for machine learning applications. To address this gap, we propose a structured classification framework for physics problems at the elementary school level (Shamshin, 2024; de Souza et al., 2024). Our approach focuses on the systematic annotation and organization of tasks with pedagogically and cognitively relevant features, creating a dataset suitable for training and evaluating AI models in education. By classifying physics problems based on problem type, cognitive complexity, physical quantities, and other key attributes, this framework supports the development of intelligent educational systems capable of adaptive task recommendation, personalized learning, and formative assessment. The problems were processed and annotated according to relevant criteria, including problem type (conceptual, quantitative, mixed), cognitive complexity level according to revised Bloom’s taxonomy (Krathwohl, 2002), number of physical quantities and formulas, key concepts, and measurement units (Table 1). Problem complexity is determined by the number of reasoning steps required, from simple calculations (e.g., finding speed using 𝑣 = 𝑠/𝑡) to multi-concept problems (e.g., calculating acceleration while considering friction and inclined forces). For instance, determining a box's acceleration when pulled at an angle requires resolving forces, calculating friction, and applying Newton's laws - a 5-step analysis typical for 8th grade physics. The dataset covers key physics topics for grades 7 and 8, such as force and motion, oscillations, and optics.