Evolutionary clustering is a powerful technique for detecting dynamic communities in temporal graphs, where network structures change over time. Unlike static clustering methods, evolutionary clustering incorporates temporal smoothness to ensure consistency between consecutive time snapshots while accurately capturing evolving communities. This approach is essential for various real-world applications, such as temporal social and biological network analysis. In this paper, we present a robust evolutionary clustering method based on non-negative matrix factorization (NMF) to detect dynamic communities. Our approach derives a low-rank approximation of the temporal network at each snapshot to remove noise and outliers. The low-rank approximation is then used to identify the community structures by integrating temporal smoothness term with non-negative matrix factorization clustering. This integration enhances the current snapshot structure by incorporating information from the previous snapshot, minimizing abrupt clustering changes and improving robustness. Experimental results on synthetic and real-world dynamic networks demonstrate that the proposed method outperforms state-of-the-art approaches in both accuracy and robustness.