Improving Knowledge Graph Embeddings through Contrastive Learning with Negative Statements