Knowledge graph embedding techniques are widely used for knowledge graph refinement tasks such as graph completion and triple classification. These techniques aim at embedding the entities and relations of a Knowledge Graph (KG) in a low dimensional continuous feature space. They do so by learning a scoring function that computes the plausibility of a fact. Hence the evaluation of a refinement task is done either as a classification, or all possible substitutions are created to rank and choose the most plausible triples. In this work, we adopt a transformer-based triplet network that can create clusters based on entities or relations. By doing so, we can present an evaluation strategy that relies on very efficient spatial semantic search techniques, and that uses aggregation methods to evaluate the KG refinement tasks with less overhead. We evaluate our approach on triplet classification and relation prediction on multiple well-known benchmark knowledge graphs such as FB13, WN11, FB15K. The results are better or comparable to the state-of-the-art performance on these two refinement tasks.