Community detection is central to network analysis, but most methods target static structures. Real-world communities are often overlapping and dynamic, requiring more precise models. While methods like Clique Percolation address overlap, they typically rely on coarse time discretizations that obscure temporal patterns. The link stream (LS) model captures time-stamped interactions, with LSCPM being the only known method for overlapping community detection in this setting. In this paper, we adapt a Formal Concept Analysis-based method to the LS context, enabling high-resolution detection of dynamic, overlapping communities. We also propose tailored evaluation metrics for this temporal context.