Knowledge Space Theory as a Means of Structural–Combinatorial Analysis of Game Log Data

The analysis of learning data from digital learning environments and serious games poses significant methodological challenges. Conventional statistical methods often yield weak or non-significant results, not necessarily because learning has not occurred, but because learning processes are highly multivariate and influenced by complex interactions among learner characteristics, content, and behavior. Although artificial intelligence and machine learning can detect patterns in such data, the resulting models are often difficult to interpret and provide limited insight into learning mechanisms. This paper introduces a structural–combinatorial framework based on Competence-based Knowledge Space Theory (CbKST) for analyzing complex educational data. CbKST represents learning as structured competence states connected through prerequisite relations, enabling transparent interpretation of learner development. The framework is applied to log data from Gravitas, a serious game designed to support conceptual understanding of mechanics. Comparisons with conventional statistical analyses illustrate that CbKST provides a complementary structural perspective by reconstructing competence profiles from heterogeneous gameplay evidence that is not represented in aggregate performance measures. The results suggest that structural–combinatorial approaches can complement statistical and AI-based methods by providing more interpretable and explanatory representations of learning processes in complex digital learning environments.