Local neighborhoods play a crucial role in embedding generation in graph-based learning. It is commonly believed that nodes ought to have embeddings that resemble those of their neighbors. In this research, we try to carefully expand the concept of similarity from nearby neighborhoods to the entire graph. We provide an extension of similarity that is based on transitivity relations, which enables Graph Neural Networks (GNNs) to capture both global similarities and local similarities over the whole graph. We introduce Transitivity Graph Neural Network (TransGNN), which more than local node similarities, takes into account global similarities by distinguishing strong transitivity relations from weak ones and exploiting them. We evaluate our model over several real-world datasets and showed that it considerably improves the performance of several well-known GNN models, for tasks such as node classification. This popularity can be attributed to GNNs' adaptability and efficiency in learning from data structured as graphs, proving essential in domains where data can be naturally organized into nodes, and predictions rely on the complex relationships (edges) inter-linking these nodes. Their versatility finds applications in diverse fields such as molecular chemistry [7], social networks [8], and recommendation systems [9]. Graph Convolutional Networks (GCNs) [10], introduced by Kipf and Welling in 2017, present an efficient adaptation of Convolutional Neural Networks (CNNs) [11] for graph data. This model involves stacking layers of first-order spectral filters, succeeded by a non-linear activation function, facilitating the acquisition of graph representations [10]. Within the GNN framework, the core concept revolves around iteratively updating node states through interactions with their neighbors.