User modeling is an important area in the field of artificial intelligence. Traditionally, user modeling approaches have focused on individual features of users, leaving aside aspects of users as part of a group. However, in enterprise, academic, or political contexts, it is common to find groups of people working together to solve a collaborative task using online collaborative platforms. In this context, counting with a knowledge based on describing the users' behavior when working in a group (i.e., the collaborative profile) is vital for a system aiming at assisting them to enhance their collaborative abilities that might impact the group dynamics. Traditionally, collaborative abilities have been measured by means of questionnaires or by an external human observer. In this work, we propose an approach to build collaborative profiles by the means of observation of users' behavior in a collaborative serious game. The main advantage of our approach is that it avoids external human observers of the group dynamics and the use of self-perception questionnaires. We obtained promising results when comparing the user profiles obtained with our approach with respect to the classical measurement instruments.