Self-Regulated Learning (SRL) is a key goal of education. This set of skills is relevant especially in complex, open-ended tasks like modeling activities within Open-Ended Learning Environments. Existing SRL models are often task-specific, limiting their generalizability. This paper proposes a transferable analytical approach to assess SRL behaviors across different modeling activities. First, we construct a semantic vocabulary of typical student actions in modeling activities. We then map task-specific events onto this semantic map, creating a representation that is comparable across modeling activities, irrespective of system affordances. We evaluated the approach using OECD PISA pilot data from the 2025 Learning in the Digital World assessment, including 446 students working with two activities. Using K-means clustering on behavior frequencies, we identified two distinct clusters, labeled “Inquiry-focused” and “Solution-focused”. These groups demonstrated significantly different behavioral patterns within each activity, and these patterns were semantically similar across activities, reflecting consistent approaches to engagement in modeling activities. Finally, cluster membership was associated with performance, with “Inquiry-focused” learners consistently achieving better outcomes in both activities. These findings align with previous research, validating our approach, as bootstrapping the interaction did not hinder the ability to observe patterns. This approach offers a potentially scalable method and a more robust solution in the context of international large-scale assessments. It further enables analysis of the impact of specific affordances on SRL patterns across tasks.