Mapping Strategy Shifts with Sankey Diagrams: Insights from AI Logs in Primary CT Education

Abstract

Although AI learning companions hold promise for enhancing computational thinking education, understanding how students’ learning strategies evolve through interaction with these tools remains challenging—highlighting the need for process-oriented analysis to move beyond static outcomes and reveal how learning truly unfolds. This study applied a learning analytics approach to examine interaction logs from 122 third-grade Taiwanese students using the AI learning companion TALPer from the Taiwan Adaptive Learning Platform (TALP) for computational thinking tasks over six weeks. By coding student behaviors and visualizing strategy transitions with Sankey diagrams, we revealed the dynamic evolution of learning strategies within an AI-enhanced environment. The Sankey diagrams revealed a temporal shift in student interactions with TALPer, showing a progression from basic behaviors like information queries (16%) and language practice (24%) toward more metacognitive strategies such as planning (21%) and problem-solving (28%) as the intervention advanced, highlighting the evolving depth of engagement over time. These findings underscore the value of process-oriented learning analytics in complementing traditional assessments and offer practical insights for designing adaptive AI systems that support self-regulated learning in young students.