While cloud service providers generate automatically labels for technical attributes, the primary challenge arises in maintaining the accuracy and consistency of functional labels defined by cloud architects. The main issue stems from the fact that these labels may be incorrect or poorly formatted, potentially leading to mismanagement of cloud resources. The aim of this paper is to present an approach for unifying labels in the cloud using Natural Language Processing (NLP) techniques to establish a coherent architecture of the cloud. The key contribution of our approach is the combination of label analysis with semantic and syntactic analysis algorithms to provide complete and refined label unification. We perform an in-depth analysis of the proposed approach effectiveness using several sets of realistic multi-cloud assets deployed by cloud engineers, on which we simulated common syntactic and semantic mislabeling errors that may arise in the context of cloud resources. The experimental results confirm that our approach contributes to improving resource labels unification in multi-cloud environments compared to GPT-4o only and a configuration file based approach, thanks to the combined unification of labels.