<![CDATA[This study evaluates the questioning skills of chemistry teachers during microteaching using an AI-assisted assessment rubric. A total of 200 publicly available YouTube videos (2019–2024) were selected using defined criteria: chemistry instruction, teacher–student questioning, Indonesian language, minimum audio clarity of 45 dB, and at least 5 minutes in duration. All videos featured pre-service teachers. Transcripts were generated using Otter.ai and segmented into discrete questioning episodes. Evaluation was performed using Gemini Flash 2.0 (build: 2025.03, temperature: 0.0), a large language model configured via prompt design and anchored exemplars to assess six pedagogical indicators: question type, content relevance, question complexity, wait time, teacher’s response, and student interaction. Each indicator was rated on a 4-point scale. Reliability checks against human-coded samples (n = 40) yielded strong agreement (Cohen’s κ = 0.78). Results showed that 25% of sessions were classified as high-performing, with open-ended and cognitively demanding questions, extended wait time, and rich student engagement. In contrast, 42% were low-performing, marked by factual recall, short pauses, and minimal interaction. Clustering analysis (Gower k-medoids) identified three distinct performance profiles (average silhouette = 0.41). This AI-based framework enables reliable, scalable, and interpretable evaluation of questioning practices. A prototype feedback tool was developed, providing per-indicator scores, question examples, and suggested improvements supporting formative teacher development. Ethical compliance was ensured through the exclusive use of public, anonymized content.]]>
Copyrights © 2025
