A truss‐based approach for densest homogeneous subgraph mining in node‐attributed graphs. (13th April 2021)

• Record Type: Journal Article
• Authors: Sun, Heli; Zhang, Yawei; Jia, Xiaolin; Wang, Pei; Huang, Ruodan; Huang, Jianbin; He, Liang; Sun, Zhongbin
• Other Names: Li Ying guestEditor.; Shyamasundar R.K. guestEditor.; Wang Xinheng guestEditor.
• Is Part Of: Computational intelligence. Volume 37:Number 2(2021)
• Page Start: 995
• Page End: 1010
• Publication Date: 2021-04-13
• Subjects: attribute description -- densest subgraph -- hybridized k‐truss -- Jaccard support -- node‐attributed graph; Artificial intelligence -- Periodicals; Computational linguistics -- Periodicals; 006.3
• DOI: 10.1111/coin.12448
• Languages: English
• ISSNs: 0824-7935

Abstract

Abstract: In a wide range of graph analysis tasks such as community detection and event detection, densest subgraph mining is important and primitive. With the development of social network, densest subgraph mining not only need to consider the structural data but also the attributes information, which descripts the features of nodes or edges. However, there are few researches on densest subgraph mining with attribute description. In this article, we only focus on the node‐attributed graph. According to the properties of structure and attribute in node‐attributed graphs, we define a novel dense subgraph pattern, called hybridized k ‐truss in attribute‐augmented graph. A hybridized k ‐truss is a subgraph that consists of structural nodes and attribute nodes, of which there are at least ( k − 2) common neighbors between any two connected nodes. We introduce the densest hybridized truss problem, and the densest hybridized truss mapping to a densely connected subgraph with homogenous attributes in the original graph. We propose a densest hybridized truss extraction (DHTE) algorithm for node‐attributed graphs, to automatically find the densest subgraph with high density and homogenous attributes at the same time. Extensive experimental results of 21 real world datasets demonstrate the effectiveness and efficiency of DHTE over state‐of‐the‐art methods, through comparison about structural cohesiveness and attributive homogeneity.