Detecting communities in attributed networks poses a considerable challenge in network science, requiring advanced techniques to effectively merge structural and attribute data. In complex network analysis, focusing solely on direct connections (first-order relationships) can be restrictive. Implicit connections, which take into account higher-order relationships, reveal hidden links between individuals who may not be directly connected, offering a more comprehensive understanding of the network's structure. Spectral Clustering is a key algorithm for community detection, capable of identifying groups within a complex network and it performs well. It can also be combined with other algorithms for a more flexible and adaptable approach. However, traditional spectral clustering methods often neglect node attributes, concentrating mainly on first-order connections of the network structure, which can result in incomplete community detection. This thesis presents a novel spectral clustering approach enhanced by Pointwise Mutual Information (PMI) to improve the accuracy and relevance of community detection in attributed networks. Our approach constructs an enhanced similarity matrix incorporating PMI to capture both direct and indirect node connections. We use Depth-First Search (DFS) to traverse the network, enriching the structural matrix with indirect relationships. Additionally, pairwise constraints 'Must-Link' and 'Cannot-Link' are included to further refine community detection, ensuring the clustering process respects inherent data relationships. The node attribute matrix is then combined with the enhanced structural matrix using double auto encoder is one of the effective techniques. This integration aims to utilize both structural and attribute information, resulting in more precise and meaningful community detection. Empirical validation on various real-world datasets shows that our proposed method significantly outperforms other community detection algorithms, providing a robust framework for analyzing complex attributed networks.
