王科植Wang, Ko-Chih李美娉Josephine, Nathania2025-12-092025-08-072025https://etds.lib.ntnu.edu.tw/thesis/detail/ca14fb40aaad5d78950bff645c73803c/http://rportal.lib.ntnu.edu.tw/handle/20.500.12235/125817NoneProviding effective feedback is critical for developing visualization construction literacy. Although computer-based feedback systems are able to help instructors manage large classes, adaptive mechanisms that tailor delivery to individual needs remain underexplored in visualization construction literacy. In this study, we investigate the impact of adaptive feedback, by a combination of rule-based progression integrated to a large language model (LLM), versus non-adaptive feedback in improving visualization construction literacy. We create a system that analyzes student's visualization specifications, detects errors, and delivers feedback. The non-adaptive condition errors are presented directly, while the adaptive condition includes processing input history and uses an LLM to generate personalized feedback that adjusts in granularity and sequence over time. A user experiment involving 30 students who are beginners in visualization demonstrated that both adaptive and non-adaptive feedback significantly improved students’ visualization construction literacy. However, adaptive feedback led to greater overall improvement and showed more consistent effects across different task types. Students' preferences for adaptive feedback methods vary, highlighting the need for adaptive systems to balance educational effectiveness with user learning style. These findings suggest that adaptive feedback delivery powered by an LLM has a strong potential to improve visualization construction literacy, particularly when tailored to student's individual preferences and learning contexts.NoneVisualization in EducationVisualization Construction LiteracyAdaptive FeedbackLarge Language Model學術論文