Due to the growing complexity of projects, their risks have increased in number and criticality. Risk lists thus need to be broken down into smaller, more manageable clusters. Classical clustering techniques are generally based on a single parameter, like risk nature, criticality or ownership. Risk interactions are therefore not properly considered when building up clusters. That is why this paper aims at grouping risks so that the com-munication and coordination between the actors who are committed in the management of the project and its risks are facilitated. The work is based on an optimization algorithm which maximizes interaction rate within the risk clusters. This paper focuses on two additional points. First, the optimization problem formulation is enriched by some constraints related to the risk owners, not only to the risks. Second, a frequency approach is introduced, to test different configurations, in order to improve the robustness of the clustering decision. It enables meaningful and operationally realistic actors groups to be organized, regarding not only the interaction rate between risks but also the relationships between risk owners. Our clustering approach encourages people to meet together and communicate/ coordinate better, which we hope will contribute to prevent some unde-sired complex phenomena