The field of medicine has experienced rapid advancements, accumulating a vast quantity of medical literature and clinical notes. However, a common challenge in automated medical language processing arises from multiple expressions for many medical terms, resulting in either multiple meanings assigned to a single term or multiple terms referring to a single meaning. Addressing this challenge, therefore, requires the development of efficient models for the normalization of specialized terms. In this research paper, we propose a novel method of graph neural network (GNN) in conjunction with a recommendation algorithm to explore the intricate relationships among words and sentences. This combined approach aims to enhance the effectiveness of clinical terminology normalization and resolve the issue of polysemy. Specifically, we incorporate GraphSAGE with a recommendation algorithm to tackle the task of word sense disambiguation. Our experiments demonstrate that integrating a graph neural network and recommendation algorithm for word sense disambiguation yields a noteworthy average Micro F1 score of 64.6%, representing a significant improvement compared to other classical models.