Improving both the quantity and quality of existing data are placed at the center of research for adaptive micro open learning. To cover this research gap, our work targets on the current scarcity of both data and rules that represent open learning activities. An evolutionary rule generator is constructed, which consists of an outer loop and an inner loop. The outer loop runs a genetic algorithm (GA) to produce association rules that can be effective in the micro open learning scenario from a small amount of available data sources; while the inner loop optimizes generated candidates by taking into account both rare and negative association rules (NARs). These optimized rules are further applied in refining and augmenting data denoting learners’ behaviors in open learning into a low‐dimensional, descriptive and interpretable form. The performance of rule discovery and data processing have been empirically evaluated using genuine open learning data.