A novel network core structure extraction algorithm utilized variational autoencoder for community detection. (15th July 2023)

• Record Type: Journal Article
• Authors: Fei, Rong; Wan, Yuxin; Hu, Bo; Li, Aimin; Li, Qian
• Is Part Of: Expert systems with applications. Volume 222(2023)
• Publication Date: 2023-07-15
• Subjects: Community detection -- Complex networks -- Core structure -- Variational autoencoder -- Clustering; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2023.119775
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Community detection technologies have the general research significance in complex networks, in which the topology information of network is worthy to be the focus for its widely application. It is the definition of community structure that the connection of nodes in the community is dense with the connection of nodes outside the community is sparse, which is corresponding to the core structure in the complex real networks is represented by a compact and dense set of connected nodes. While all the notes in the network are considered by the traditional topology, it is hard to extract the core structure with the continuous, exponential growth of community networks. In this paper, a novel network core structure extraction algorithm utilized variational autoencoder for community detection(CSEA) is proposed for finding the community structure more accurately. Firstly, the K-truss algorithm is used to find the core structure information in the network, and the similarity matrix is generated by similarity mapping combined with local information. Secondly, the variational autoencoder is used to extract and reduce the dimension of the similarity matrix containing the core structure of the network, and the low-dimensional feature matrix is obtained. Finally, the K-means clustering algorithm is utilized to obtain the community structure information. We compare CSEA algorithm with 18 different types of community detection algorithms using 4 evaluation metrics on 19 complex real networks. By extensively evaluating our algorithm on large real-world datasets, we show that CSEA algorithm has an excellent community division effect in dense complex real networks, especially in small and medium-sized networks, and it can accurately divide the complex real networks with unknown community structure. Simultaneously, CSEA algorithm also reveals some efficiency advantage in its on-line test. Highlights: Core structure in complex networks is meaningful for community detection. A new variational autoencoder model is designed to speed up the clustering process. A novel structure is proposed combines the logic analysis with neural networks.