Knowledge Graphs (KGs) serve as a fundamental resource for structured knowledge, providing a cornerstone for numerous analytical endeavors, particularly within the domains of machine learning, data mining, and artificial intelligence research. The domain of Knowledge Graph Completion (KGC) is focused on the intellectual pursuit of inferring missing (latent) relational connections from existing (observed) data points, thereby enhancing the completeness and accuracy of the KG’s informational fabric. A significant body of contemporary research has emphasized the utility of Graph Convolutional Networks (GCNs) for KGC task. These lines of models have demonstrated a prowess in representation learning for graph-structured data, yielding promising outcomes in the field. Wherein, the aggregation procedure plays a pivotal role in GCN-based KGC models, however it frequently aggregates unreliable or irrelevant neighbor node information, which can compromise the accuracy of reasoning outcomes. To overcome this problem, this work proposes an optimization method respect to the aggregation procedure for GCN-based KGC models, by leveraging analogy reasoning from both entity-level and relation-level. Especially, it helps to introduce beneficial signals from topologically-indirect however semantically-analogical neighboring domain, beyond conventional topologically-direct neighboring domain. Besides, this work could be viewed as a versatile and readily integrable plug-in unit, designed to be interoperable with existing GCN-based KGC frameworks.