<p>Automatically extracting knowledge from diverse datasets is a valuable task that helps experts explore new types of data while reducing the time spent on manual annotations. This is particularly important for emerging fields such as emergency management and environmental monitoring. Traditional unsupervised methods often struggle to capture experts’ intuitions or integrate non-formalized knowledge. On the other hand, supervised methods typically require a substantial amount of prior knowledge to function effectively. Constrained clustering, a semi-supervised approach, addresses these challenges by allowing experts to incorporate their knowledge into the clustering process. However, it often yields suboptimal results because it is difficult for experts to provide constraints that are both informative and coherent. Building on the idea that it is easier to critique than to construct, this article introduces a novel method called <span>I-Samarah</span>, an incremental constrained clustering approach. This method alternates between a clustering phase, where expert-provided constraints are applied, and a critique phase, where experts provide feedback on the clustering results. Through an iterative process, the method refines the clusters, improving their alignment with expert knowledge. We empirically demonstrate the effectiveness of <span>I-Samarah</span> using remote sensing image time series, comparing it to other constrained clustering methods in terms of result quality and to supervised methods in terms of annotation efficiency.</p>