Monitoring and interpreting sequential user activities contributes to enhanced, more fine-grained user models in e-learning systems. We present in this paper different behavioural patterns from the domain of problem-solving that can be determined by targeted, ultimately automated clustering. For the identification of these patterns, we apply a new approach - based on the modeling of activity sequences - to real-world learning activity sequence data, monitored via an Intelligent Tutoring System. This paper describes the identified behavioural patterns, explains the process used for their detection, and compares the patterns to related ones in earlier literature. It further discusses implications of the patterns themselves, and of the employed approach, on adaptively supporting individual and group-based collaborative learning.