Knowledge Graph Completion (KGC) is a challenging problem that aims to infer and learn from existing partial data in order to discover implicit relationships within the knowledge graphConvolutional networks (GCN) are an effective structure in Graph Neural Networks and used in KG representation learning and inference in recent years. However, GCNs have two limitations when dealing with large-scale sparse-relation KGs in the traditional Cartesian coordinate system. Firstly, it fails to model the semantic hierarchy relationships between entities. Secondly, it has high computational complexity when handling large-scale KGs. To overcome these limitations, this paper proposes a novel approach for Sparse-Relation Knowledge Graph Completion (SRKGC) named Multi-Circular Graph Convolutional Networks (CGCN). CGCN consists of a Multi-Circular Encoder (modeling entity and relation representations through polar coordinate systems), a Modulus and Phase Updater (enhancing entity and relation representations by aggregating neighbor information), and a 2D Graph Convolutional Decoder (extracting key features through convolution from learned embeddings). The code can be found here: https://github.com/YuZhan/CGCN .