Community detection in graphs is a fundamental task in network analysis, aimed at identifying clusters of densely connected nodes. It has applications in many fields, such as social media, biological systems and cybersecurity. While traditional methods focus on static graphs, many real-world networks are dynamic, with evolving structures and community memberships. This thesis addresses the challenge of dynamic community detection, proposing a semi-supervised approach. In this context, semi-supervision means leveraging both graph structure and knowledge of part of ground truth communities in order to infer those in the remaining part of the graph. Therefore, in a dynamic setting, we have a graph evolving for many timesteps and, in each time instant, we know part of the ground-truth communities.