DrawSim-PD: Simulating Student Science Drawings to Support NGSS-Aligned Teacher Diagnostic Reasoning

TL;DR

DrawSim-PD tackles the privacy barrier to sharing authentic student drawings for NGSS-aligned teacher PD by introducing capability profiles that guide the joint generation of a student-like drawing, a first-person reasoning narrative, and a teacher-facing diagnostic concept map. The framework comprises three modules—NGSS-aligned student simulation, drawing-centric synthesis, and diagnostic concept mapping—operating on a shared, four-level capability profile across 100 NGSS topics to produce 10,000 artifacts with structured metadata. An expert feasibility study reports strong NGSS alignment and plausible grade-band differentiation, with cross-modal coherence validated via quantitative (CLIP) metrics and qualitative teacher feedback, while ablation shows capability profiles are essential for differentiating performance levels. The resulting open corpus and infrastructure enable scalable, privacy-preserving calibration, targeted misconception libraries, and robust visual-assessment research, with potential extensions to adaptive calibration and explainable diagnostic tooling.

Abstract

Developing expertise in diagnostic reasoning requires practice with diverse student artifacts, yet privacy regulations prohibit sharing authentic student work for teacher professional development (PD) at scale. We present DrawSim-PD, the first generative framework that simulates NGSS-aligned, student-like science drawings exhibiting controllable pedagogical imperfections to support teacher training. Central to our approach are apability profiles--structured cognitive states encoding what students at each performance level can and cannot yet demonstrate. These profiles ensure cross-modal coherence across generated outputs: (i) a student-like drawing, (ii) a first-person reasoning narrative, and (iii) a teacher-facing diagnostic concept map. Using 100 curated NGSS topics spanning K-12, we construct a corpus of 10,000 systematically structured artifacts. Through an expert-based feasibility evaluation, K--12 science educators verified the artifacts' alignment with NGSS expectations (>84% positive on core items) and utility for interpreting student thinking, while identifying refinement opportunities for grade-band extremes. We release this open infrastructure to overcome data scarcity barriers in visual assessment research.

Figures (13)

• Figure 1: The DrawSim-PD Framework. (Left) Teachers struggle to practice diagnostic reasoning due to a scarcity of privacy-compliant student drawings. (Center) The system uses capability profiles to bridge the gap, generating student-like artifacts with controlled misconceptions. (Right) The output serves as scalable infrastructure for teacher calibration and professional development.
• Figure 2: The DrawSim-PD framework comprises three modules: (1) NGSS-Aligned Student Simulator, which generates topic-specific evidence statements and capability profiles representing diverse K--12 student performance levels; (2) Drawing-Centric Synthesis, which produces reasoning narratives and student-like drawings conditioned on capability profiles; and (3) Diagnostic Concept Mapping, which converts outputs into a four-layer concept map linking observable features to underlying understanding and suggested instructional next steps.
• Figure 3: NGSS-Aligned Student Simulator. This module converts NGSS performance expectations (Science and Engineering Practices, Disciplinary Core Ideas, Crosscutting Concepts) into capability profiles encoding what students at each performance level (Emergent, Developing, Proficient, Advanced) can and cannot demonstrate.
• Figure 4: Drawing-Centric Synthesis. This module generates student-like drawings by conditioning text-to-image generation on capability profiles, coordinating reasoning narratives with visual outputs to maintain coherence across modalities.
• Figure 5: Example DrawSim-PD output. Given topic, NGSS code, grade, and performance level, the framework generates: (a) a student-like scientific illustration, (b) a diagnostic concept map linking visual features to underlying reasoning, and (c) reference evidence statements defining targeted learning goals.