The knot model is aimed at obtaining a trust-based reputation in communities of strangers. It identifies groups of trustees, denoted as knots and among whom overall trust is strong, and is thus considered the most capable solution for providing reputation information to other members within the same knot. The problem of identifying knots in a trust network is modeled as a graph clustering problem. When considering dynamic and large-scale communities, the task of keeping the clustering correct over time is a great challenge. This paper introduces a clustering maintenance algorithm based on the properties of knots of trust. A maintenance strategy is defined that addresses violations of knot properties due to changes in trust relations that occur with time in response to the dynamic nature of the community. Based on this strategy, a reputation management procedure is implemented in two phases: the first identifies the essence of change and makes a decision regarding the need to improve knot clustering. The second phase locally modifies the clustering to preserve a stable network structure while keeping the network correctly clustered with respect to the knot utility function. We demonstrate by simulation the efficiency of the maintenance algorithm in preserving knots quality, for cases in which only local changes have occurred, to ensure the reliability of the reputation system.