Knowledge Graph Embedding (KGE) is crucial for representing entities and relations in vector space to support downstream tasks. Although neural network-based models, especially CNNs, have shown promise, they still struggle to capture complex interactions and semantic information from neighborhood structures. This paper introduces InsE, a novel KGE model that integrates an Adaptive Interaction-Aware Learning Module and a Neighbourhood Semantic Extraction Module to enhance entity-relation interaction and neighborhood semantic extraction. InsE adopts a cascading architecture comprising channel attention and normalization layers, coupled with a Glocal Triple Scoring mechanism, to enhance embedding learning. Extensive experiments on benchmark datasets demonstrate that InsE achieves competitive performance with high parameter efficiency, especially excelling on dense KGs.