Knowledge Graph Completion (KGC) seeks to infer missing triplets in Knowledge Graphs (KGs), thereby mitigating their inherent incompleteness. Existing methods typically focus on embedding some single features, including relation types or entity description knowledge. But they fall short in fully leveraging knowledge graph features and external semantic knowledge. To overcome these limi-tations, we propose a knowledge-aware reasoning model (ECHO), which designs a novel multi-source knowledge representation learning mechanism for missing relation prediction. Our model develops a tailored embedding space through dual parallel components: 1) Semantic Entity En-hancement (SEE), which refines the triplets’ semantic representation by continuously pre-training the BERT model; 2) Topological Relation Yielding (TRY), which simultaneously performs contextual message passing and relational path aggregation to capture the connectivity from the head to the tail entity. The experimental results show that ECHO significantly outperforms other baselines on four da-tasets, achieving over 7.9% H@1 improvement and 5.6% MRR improvement on the NELL995 dataset. The implementation details will be open on GitHub.