The study of social networks has received a lot of attention lately. One way to represent social networks is via a graph. Every person is referred to as a node and the links that connect them as edges when analyzing social networks. In real-world network graphs, it is challenging to identify communities and describe the relationships between items and persons. There are several well-established techniques for finding the connected nodes that lead to the identification of communities. This research integrates the k-nearest neighbours (KNN) algorithm with unnormalized spectral clustering to provide a unique approach for community recognition in social networks. The method makes use of KNN for local neighbourhood refining and spectral clustering for global structure analysis. The primary objective of the algorithm proposed in this study is to identify and eliminate any noisy nodes from the identified communities, hence enhancing the quality of the identified communities. Experimental results on real-world datasets demonstrate the method's effectiveness in accurately identifying community structures.