Capturing evolving community structures in social networks has recently become a focal point, especially for its practical applications in social media. Existing models often assume network structures evolve monotonically, disregarding the patterned nature of real-world network changes. Extracting valuable knowledge from previous time snapshots is challenging, and using inappropriate knowledge may negatively impact current snapshots. Thus, assessing and selecting high-quality knowledge remain critical tasks. This paper introduces a metric, Jaccard Similarity based on Motifs (JSM), for evaluating knowledge quality. Additionally, we propose a straightforward yet effective method for knowledge selection using this metric. Experimental results demonstrate the superiority of our approach over existing dynamic community detection algorithms.