Recent technological advancements have substantially advanced research in knowledge graph completion. Nevertheless, existing methodologies encounter challenges in effectively capturing discriminative data features and processing them appropriately under high data similarity scenarios. This study focuses on enhancing feature learning mechanisms in graph neural network models. Specifically, we conduct systematic investigations through the refinement of the GraIL framework. A clustering-based negative sampling strategy is proposed to enhance the model’s discriminative capability. Furthermore, the construction of analogical samples enables effective extraction of latent feature similarities, while the implementation of analogical loss functions facilitates comprehensive feature information enhancement. The integration of analogical reasoning methodology yields significant performance improvements in the proposed AnaGraIL model for knowledge graph completion tasks. Extensive experiments conducted on WN18RR, FB15k-237, and NELL-995 datasets demonstrate consistent enhancements in both AUC-PR and Hits@10 evaluation metrics.