<p>A Social Network (SN) is often described as a graph with nodes representing individuals or things and edges denoting connections or relationships. Social Network Analysis (SNA) characterizes its underlying structural and connectivity patterns, which influence information flow and community formation. Community detection (CD) discovers the group of nodes that are more strongly connected in the network by revealing the hidden structure beneath it. Despite the widespread use of traditional and Deep Learning (DL) methods for CD, these approaches often face limitations, such as scalability issues, the need for manual parameter tuning, reduced accuracy for complex community structures, difficulties in managing network complexities and particularly challenges in identifying overlapping communities. In order to overcome the limitations of identifying overlapping communities in complex systems, this paper has proposed an approach integrating Variational Graph Autoencoders (VGAE) and the Clique Percolation Method (CPM). The VGAE components learn the low-dimensional embeddings that are then used in CPM to identify overlapping communities based on local cliques. Our approach, when applied to the Karate Club dataset, achieved an Omega Index score of 0.7718, significantly outperforming the existing models. To prove the efficiency of the proposed work, further evaluations have been done on the quality of embedding using AUC, AP, Trustworthiness along with the structure of the datasets. For evaluating the semantic quality, we have used ONMI. The complex structured datasets like Citeseer, Cora, and DBLP has also been evaluated which yields Omega Index scores of 0.8224, 0.8207, and 0.8304, respectively.</p>