Nested knowledge graphs (NKGs), where subjects or objects can themselves be triples, encode higher-order relationships that traditional knowledge graph embedding (KGE) methods struggle to capture. These nested structures introduce complexity in reasoning tasks like link prediction, as they require handling relationships between triples. We propose HOKE, a hybrid embedding model designed specifically for NKGs to address this. HOKE utilizes a three-layer architecture: local embeddings capture the semantics of original KG triples; structural embeddings refine these by emphasizing shared elements between triples via a line graph representation; and nested embeddings encode higher-order relationships using a nested line graph, where triples are treated as nodes. This approach preserves both local semantic meaning and global structural dependencies. Additionally, HOKE introduces a task-driven subgraph extraction mechanism that enhances reasoning efficiency by focusing only on relevant portions of the NKG. HOKE bridges the gap in existing KGE models, providing a scalable solution to handle the complex relational structures inherent in large-scale NKGs.