A Service Clustering Based on GCN Unsupervised Community Detection

Introduction

Service computing, as a new generation of distributed computing architecture, has been widely used in many fields. Web services in particular are rarely isolated; instead, they often exist in the form of integrated or composite services (Kolberg et al., 2023). As a result, the complex interactions among web services form a large and intricate system driven by user requirements. A software system based on web services can be conceptualized as a complex network in which web services serve as nodes with rich attributes, and the collaborative relationships between them are represented as edges, signifying direct or indirect connections. How to analyze and manage such a complex service network has become an urgent problem.

Many complex systems in the real world can be abstracted as a network (Jannesari et al., 2024), such as social networks (Adamic et al., 2005), biological networks (Girvan et al., 2002), citation networks (Sen et al., 2008), polycrystalline systems (Deng et al., 2020), and service networks (Deng et al., 2024). In these networks, the nodes are often associated with a variety of attributes. A notable feature of many networks is the presence of community structures (Perozzi et al., 2014)—the organization of nodes into groups, whereby nodes within the same group are densely interconnected or share similar attributes. Community detection plays a crucial role in uncovering the mesoscale properties of complex networks (Deng et al., 2024). For instance, it can assist in identifying protein complexes and functional modules in protein–protein interaction networks.

Community structure (Perozziet al., 2014) is one of the most important and fundamental topological properties in complex networks. Web service–based software systems typically consist of interactive web services driven by user requirements. To fulfill these needs, the system must select multiple reliable web services that interact to achieve the target task. Any web service related to a specific business must have a direct or indirect relationship with others. These relationships in a service network can naturally form communities. Thus, web services can be grouped into service communities on the basis of their interactions. Each community reflects the themes and functions of the services, which can enhance the quality of service discovery and selection.

Many studies have explored community structure from a global perspective, treating the entire network as a whole and optimizing a global quality function. These methods are computationally efficient because they do not require analysis of the entire network. Among them, local expansion methods are widely used for local community detection in large networks. These methods build local communities around specified seed nodes by greedily adding nodes until a local optimum of a quality function is reached. However, such methods require prespecified seed nodes, and the results are sensitive to the initial seeds (Danon et al., 2005). To address this issue, Wang et al. (2021), inspired by Rodriguez et al. (2014), proposed a method to automatically identify structure centers in a network. These centers are characterized by higher local density than their neighbors and a relatively large distance from other nodes with higher densities, making them suitable as seed nodes for local expansion methods.