An Attributed Graph Clustering Approach Based on Nearest Neighbor Graph

The traditional attribute graph clustering approaches suffer from low computation efficiency and poor accuracy problems. How to design a general attribute graph cluster method, which leverages both the attribute and structure characteristics, has become a new challenge. This paper proposes an attribute graph clustering algorithm based on $k$ Nearest Neighbor(kNN) graph, called KAGC, which leverages both the traditional structured graph clustering method SCAN and kNN graph. KAGC performs a three-step fusion. Firstly, KAGC constructs a weighted graph based on the mixed similarity of attribute and structure, then clusters the nodes based on high similarity edges in DFS order, and finally merges the small clusters by joint strength. Experimental results demonstrate that the KAGC can effectively cluster the attribute graph and better balance the attribute and structural information of nodes.