Assessing ChatGPT-generated comments for video-based learning content to enhance knowledge and emotional support based on scaffolding theory

Texto completo

With the rise of Generative Artificial Intelligence (GenAI), tools such as ChatGPT have shown great potential in enhancing online learning. However, existing applications in video-based learning remain limited, as they typically require learners to pause or switch interfaces to access AI support, disrupting learning continuity. To address this challenge, this study proposes a framework that utilizes ChatGPT to generate comments for an individual learner (i-Comments), displayed as scrolling text over educational videos, providing both knowledge and emotional support based on scaffolding theory. The present study evaluates the framework's generation capabilities to gain empirical insight that will inform future individualization based on learner profiles. The proposed framework integrates multimodal video information analysis, individual information and comment attribute settings, generation of i-Comments, and customized display, through which pedagogical principles such as cognitive load theory, the zone of proximal development, and scaffolding theory are systematically translated into technical components, including few-shot learning, prompt engineering, and entropy-based density control. A validation experiment compared ChatGPT-generated and human-generated comments across linguistic and pedagogical dimensions, including part-of-speech composition, 3-gram diversity, Zipf's law conformity, readability, and topical relevance. Results indicate that ChatGPT-generated comments are more complex and descriptive, whereas human comments show stronger topic relevance and natural language norms, leading to higher perceived helpfulness under the GPT-3.5 condition used in the main comparison; a supplementary ablation with a newer model (GPT-5.4) suggests that this gap narrows substantially. This research provides foundational knowledge for developing Agentic AI supported educational systems by empirically evaluating and exploring key capabilities required for autonomous instructional support, offering insights into the feasibility of applying generative AI to enhance learner engagement in video-based learning.